chore(release): prepare v3.8.0

Co-authored-by: Codex <noreply@openai.com>

.agents/skills/github-pr/SKILL.md

@@ -83,6 +83,9 @@ gh pr create --repo "$pr_repo" --base "$pr_base" --head "$pr_head" \
    按上述核对流程收集一手证据，区分原有问题、环境或配置差异与 PR 新引入的问题，说明主分支已覆盖的部分。
 3. 从官方方案评估解决方向：说明官方如何处理同一场景，检查 PR 是否遵循其协议语义和职责归属、
    是否复用已有能力或正确传递官方合同；核对偏离官方方案的理由与运行证据。
+   审查 OpenAI 官方上游与 Codex 的对接时，默认官方返回与 Codex 兼容、直接对接正常，重点检查本项目是否误解、
+   改写或破坏官方合同。不得仅凭自行构造的异常上游数据揣测官方不兼容，也不据此报告缺陷、提出改进建议，
+   或要求增加防御分支与测试。讨论特定版本或场景的对接例外须有真实响应或实际复现证据，并区分代理链路与官方直连。
 4. 按贡献文档的审查维度追踪完整调用链，找到共享业务事实、同类实现及已有验证。
    对既有特殊分支查清服务的场景，再判断本次改动是否引入或加重问题。
    主动检查本次变更新引入或加重的冗余与代码异味，说明具体位置、维护成本或行为影响及最小修复方向；

README.md

@@ -72,6 +72,7 @@ curl -i http://127.0.0.1:8080/healthz
 
 健康检查返回 `204 No Content` 后，打开 `http://127.0.0.1:8080`，
 使用 `admin@cpr.local` 和刚设置的管理员密码登录。
+API Key 持有者可在同一登录页切换登录身份，进入 `/key-usage` 查看自己的用量、趋势、请求日志、额度与健康时间线；不能访问管理员页面。
 
 默认地址只能在服务器本机访问。从其他设备使用时，需要配置
 [HTTPS 反向代理](deploy/README.md#公网访问)。

backend/apps/gateway/src/bootstrap.rs

@@ -14,6 +14,8 @@ pub struct GatewayConfig {
     host: HostConfig,
     store: gateway_store::StoreConfig,
     admin: gateway_admin::AdminConfig,
+    #[serde(default)]
+    client: gateway_admin::ClientConfig,
     api: gateway_api::ApiConfig,
     openai: provider_openai::OpenAiConfig,
     xai: provider_xai::XaiConfig,
@@ -34,6 +36,9 @@ impl LoadableConfig for GatewayConfig {
         self.admin
             .resolve_and_validate(source_dir)
             .map_err(|_| ConfigError::InvalidField("admin"))?;
+        self.client
+            .resolve_and_validate(source_dir)
+            .map_err(|_| ConfigError::InvalidField("client"))?;
         self.api
             .resolve_and_validate(source_dir)
             .map_err(|_| ConfigError::InvalidField("api"))?;
@@ -55,6 +60,7 @@ pub async fn run() -> Result<(), BootstrapError> {
         host,
         store,
         admin,
+        client,
         api,
         openai,
         xai,
@@ -75,15 +81,17 @@ pub async fn run() -> Result<(), BootstrapError> {
     host.report_startup_ready("Core");
     let mut admin = gateway_admin::initialize(
         admin,
+        client,
         store.admin_ports(),
-        vec![openai.admin_provider(), xai.admin_provider()],
-        core.snapshot_control(),
-        (
-            core.account_probe(),
-            host.proxy_probe(provider_openai::build_reqwest_client_with_custom_ca),
-        ),
-        host.client_distribution_resolver(),
-        host.system_operations(),
+        gateway_admin::AdminRuntimePorts {
+            providers: vec![openai.admin_provider(), xai.admin_provider()],
+            snapshot: core.snapshot_control(),
+            account_probe: core.account_probe(),
+            proxy_probe: host.proxy_probe(provider_openai::build_reqwest_client_with_custom_ca),
+            client_distribution: host.client_distribution_resolver(),
+            system: host.system_operations(),
+            client_key_verifier: core.client_key_verifier(),
+        },
     )
     .await?;
     host.report_startup_ready("Admin");

backend/apps/gateway/tests/architecture.rs

@@ -70,6 +70,8 @@ fn core_value_owners_do_not_depend_on_execution_or_routing() {
             "identity.rs" => Some(&["validation"]),
             "upstream.rs" => Some(&["validation"]),
             "event.rs" => Some(&["metering", "operation", "upstream", "validation"]),
+            "account/selection.rs" => Some(&["account", "concurrency", "identity", "validation"]),
+            "concurrency.rs" => Some(&["error"]),
             "account/store.rs" => Some(&["account", "error", "identity", "validation"]),
             path if path.starts_with("policy/") => Some(&["account", "policy", "validation"]),
             path if path.starts_with("account/") => Some(&["account", "identity", "validation"]),
@@ -176,7 +178,7 @@ const PACKAGE_TO_MEMBER: &[(&str, &str)] = &[
 
 /// Adapter/provider 根门面的稳定合同模块；任何增减都必须同步完成边界审计。
 const ADAPTER_PUBLIC_MODULES: &[(&str, &[&str])] = &[
-    ("crates/gateway-api", &["admin", "openai"]),
+    ("crates/gateway-api", &["admin", "auth", "openai"]),
     (
         "crates/gateway-host",
         &[

backend/apps/gateway/tests/bootstrap.rs

@@ -278,6 +278,37 @@ fn config_loader_should_reject_admin_password_with_compose_interpolation() {
     assert_rejected(valid_config().replace(ADMIN_PASSWORD, "unsafe$password"));
 }
 
+#[test]
+fn config_loader_should_reject_zero_client_session_ttl() {
+    let mut config = valid_config_document();
+    *config
+        .pointer_mut("/client/session_ttl_minutes")
+        .expect("example client session TTL") = serde_json::json!(0);
+    assert_rejected(config.to_string());
+}
+
+#[test]
+fn config_loader_should_default_missing_client_section() {
+    let mut config = valid_config_document();
+    config
+        .as_object_mut()
+        .expect("example config mapping")
+        .remove("client")
+        .expect("example client section");
+    parse_config(&config.to_string()).expect("client defaults when the section is omitted");
+}
+
+#[test]
+fn config_loader_should_reject_missing_client_session_ttl() {
+    let mut config = valid_config_document();
+    config["client"]
+        .as_object_mut()
+        .expect("example client mapping")
+        .remove("session_ttl_minutes")
+        .expect("example client session TTL");
+    assert_rejected(config.to_string());
+}
+
 #[test]
 fn config_loader_should_reject_removed_fingerprint_section() {
     assert_rejected(valid_config().replace(
@@ -365,6 +396,18 @@ fn valid_config() -> String {
         )
 }
 
+fn valid_config_document() -> serde_json::Value {
+    // 按字段修改样例，避免注释或排版变化让测试输入悄悄失效；JSON 仍可由 YAML 文件入口加载。
+    config::Config::builder()
+        .add_source(config::File::from_str(
+            &valid_config(),
+            config::FileFormat::Yaml,
+        ))
+        .build()
+        .and_then(config::Config::try_deserialize)
+        .expect("example config document")
+}
+
 fn parse_config(config: &str) -> Result<(GatewayConfig, tempfile::TempDir), String> {
     let directory = tempfile::tempdir().map_err(|error| error.to_string())?;
     let deploy = directory.path().join("deploy");

backend/crates/gateway-admin/src/lib.rs

@@ -5,6 +5,7 @@
 use std::{fmt, path::Path, sync::Arc, time::Duration};
 
 use gateway_core::{
+    engine::execution::ClientKeyVerifier,
     engine::probe::AccountProbe,
     routing::ProviderKind,
     runtime::SnapshotControl,
@@ -21,6 +22,8 @@ pub mod model;
 pub mod ports;
 mod use_case;
 
+pub use use_case::key_usage::KeyUsageService;
+
 pub use use_case::{
     account_groups::AccountGroupService, accounts::AccountsService, auth::AuthService,
     backup::BackupService, client_distribution::ClientDistributionService,
@@ -58,6 +61,7 @@ const WEAK_INITIAL_PASSWORDS: &[&str] = &[
 ];
 
 const BACKUP_WORKER_OWNER: &str = "backup";
+const DEFAULT_CLIENT_SESSION_TTL_MINUTES: u64 = 24 * 60;
 
 /// 只用于首次幂等创建默认管理员的启动密码。
 #[derive(Clone, Deserialize)]
@@ -98,6 +102,35 @@ pub struct AdminConfig {
     pub default_password: InitialAdminPassword,
 }
 
+/// Client 登录域的通用启动配置。
+#[derive(Debug, Clone, Deserialize, PartialEq, Eq)]
+#[serde(deny_unknown_fields)]
+pub struct ClientConfig {
+    pub session_ttl_minutes: u64,
+}
+
+impl Default for ClientConfig {
+    fn default() -> Self {
+        Self {
+            session_ttl_minutes: DEFAULT_CLIENT_SESSION_TTL_MINUTES,
+        }
+    }
+}
+
+impl ClientConfig {
+    /// 校验 Client session TTL；当前配置不含相对路径。
+    ///
+    /// # Errors
+    ///
+    /// 会话有效期为零或无法安全换算时返回错误。
+    pub fn resolve_and_validate(&mut self, _source_dir: &Path) -> Result<(), AdminConfigError> {
+        if self.session_ttl_minutes == 0 || i64::try_from(self.session_ttl_minutes).is_err() {
+            return Err(AdminConfigError::InvalidField("client.session_ttl_minutes"));
+        }
+        Ok(())
+    }
+}
+
 impl AdminConfig {
     /// 校验 Admin-owned 字段；当前配置不含相对路径。
     ///
@@ -151,6 +184,7 @@ pub enum AdminConfigError {
 pub struct AdminServices {
     proxies: Arc<dyn ProxiesService>,
     auth: Arc<dyn AuthService>,
+    key_usage: Arc<dyn KeyUsageService>,
     accounts: Arc<dyn AccountsService>,
     account_groups: Arc<dyn AccountGroupService>,
     client_keys: Arc<dyn ClientKeyService>,
@@ -164,6 +198,11 @@ pub struct AdminServices {
 }
 
 impl AdminServices {
+    #[must_use]
+    pub fn key_usage(&self) -> &dyn KeyUsageService {
+        self.key_usage.as_ref()
+    }
+
     #[must_use]
     pub fn proxies(&self) -> &dyn ProxiesService {
         self.proxies.as_ref()
@@ -243,6 +282,17 @@ impl AdminBundle {
     }
 }
 
+/// 组合根提供给控制面的运行能力；与配置和存储端口分别传入。
+pub struct AdminRuntimePorts {
+    pub providers: Vec<Arc<dyn ProviderAdmin>>,
+    pub snapshot: Arc<dyn SnapshotControl>,
+    pub account_probe: Arc<dyn AccountProbe>,
+    pub proxy_probe: Arc<dyn ports::proxy::ProxyProbe>,
+    pub client_distribution: Arc<dyn ClientDistributionResolver>,
+    pub system: Arc<dyn SystemOperations>,
+    pub client_key_verifier: Arc<dyn ClientKeyVerifier>,
+}
+
 /// 校验配置、建立动态 Provider 注册表并完成默认管理员幂等初始化。
 ///
 /// # Errors
@@ -250,17 +300,25 @@ impl AdminBundle {
 /// 配置非法、Provider 注册冲突/缺失或默认管理员初始化失败时返回错误。
 pub async fn initialize(
     mut config: AdminConfig,
+    mut client_config: ClientConfig,
     store: AdminStorePorts,
-    providers: Vec<Arc<dyn ProviderAdmin>>,
-    snapshot: Arc<dyn SnapshotControl>,
-    probes: (Arc<dyn AccountProbe>, Arc<dyn ports::proxy::ProxyProbe>),
-    client_distribution: Arc<dyn ClientDistributionResolver>,
-    system: Arc<dyn SystemOperations>,
+    runtime: AdminRuntimePorts,
 ) -> Result<AdminBundle, AdminError> {
-    let (probe, proxy_probe) = probes;
+    let AdminRuntimePorts {
+        providers,
+        snapshot,
+        account_probe: probe,
+        proxy_probe,
+        client_distribution,
+        system,
+        client_key_verifier,
+    } = runtime;
     config
         .resolve_and_validate(Path::new("."))
         .map_err(|error| AdminError::invalid(error.to_string()))?;
+    client_config
+        .resolve_and_validate(Path::new("."))
+        .map_err(|error| AdminError::invalid(error.to_string()))?;
     let registry = ProviderAdminRegistry::new(providers).map_err(map_provider_registry_error)?;
     let openai = registry
         .require(&provider_kind(OPENAI_PROVIDER_KIND)?)
@@ -272,7 +330,9 @@ pub async fn initialize(
     let auth = Arc::new(DefaultAuthService::new(
         config.default_username,
         config.session_ttl_minutes,
+        client_config.session_ttl_minutes,
         store.auth(),
+        client_key_verifier,
     ));
     auth.ensure_default_admin(config.default_password.expose())
         .await?;
@@ -296,7 +356,13 @@ pub async fn initialize(
         backup_ports.dump(),
         backup_ports.object_store(),
     );
+    let key_usage = Arc::new(use_case::key_usage::DefaultKeyUsageService::new(
+        auth.clone(),
+        store.client_keys(),
+        store.observability(),
+    ));
     let services = AdminServices {
+        key_usage,
         proxies: Arc::new(use_case::proxies::DefaultProxiesService::new(
             store.proxies(),
             proxy_probe,

backend/crates/gateway-admin/src/model/accounts.rs

@@ -18,12 +18,14 @@ pub use gateway_core::account::{
     QuotaState, resolve_account_status,
 };
 
-/// 导入时统一应用的账号调度与分组设置；缺省时保留原有导入语义。
+/// 导入时统一应用的账号备注、调度与分组设置；缺省时保留原有导入语义。
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct AccountImportSettings {
+    pub notes: Option<String>,
     pub enabled: bool,
     pub concurrency_limit: Option<AccountConcurrencyLimit>,
     pub weight: AccountWeight,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
     pub group_ids: Vec<gateway_core::routing::AccountGroupId>,
 }
 
@@ -86,6 +88,7 @@ pub struct AccountRecord {
     pub provider_kind: ProviderKind,
     pub groups: Vec<AccountGroupRef>,
     pub name: String,
+    pub notes: Option<String>,
     pub email: Option<String>,
     pub upstream_user_id: Option<String>,
     pub upstream_account_id: Option<String>,
@@ -98,6 +101,7 @@ pub struct AccountRecord {
     pub enabled: bool,
     pub concurrency_limit: Option<AccountConcurrencyLimit>,
     pub weight: AccountWeight,
+    pub model_access: gateway_core::account::AccountModelAccess,
     pub outbound_proxy: Option<gateway_core::account::OutboundProxy>,
     pub credential_state: CredentialState,
     pub credential_observed_at: DateTime<Utc>,
@@ -217,9 +221,12 @@ pub struct AccountSummary {
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct UpdateAccount {
     pub account_id: String,
+    /// 缺省保留备注；空字符串清空备注。
+    pub notes: Option<String>,
     pub enabled: bool,
     pub concurrency_limit: Option<AccountConcurrencyLimit>,
     pub weight: AccountWeight,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
     pub group_ids: Vec<gateway_core::routing::AccountGroupId>,
     pub outbound_proxy: Option<super::proxies::AccountProxySelection>,
 }
@@ -231,14 +238,15 @@ pub struct AccountUpdateResult {
     pub account_id: gateway_core::account::ProviderAccountId,
 }
 
-/// 一批账号可编辑事实的一次性替换命令。
+/// 仅修改显式字段的批量账号设置命令。
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct BatchUpdateAccounts {
     pub account_ids: Vec<String>,
-    pub enabled: bool,
-    pub concurrency_limit: Option<AccountConcurrencyLimit>,
-    pub weight: AccountWeight,
-    pub group_ids: Vec<gateway_core::routing::AccountGroupId>,
+    pub enabled: Option<bool>,
+    pub concurrency_limit: Option<Option<AccountConcurrencyLimit>>,
+    pub weight: Option<AccountWeight>,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
+    pub group_ids: Option<Vec<gateway_core::routing::AccountGroupId>>,
     pub outbound_proxy: Option<super::proxies::AccountProxySelection>,
 }
 

backend/crates/gateway-admin/src/model/auth.rs

@@ -1,6 +1,7 @@
-//! 管理员认证、会话和安全审计事实。
+//! 控制面认证、统一会话和管理员安全审计事实。
 
 use chrono::{DateTime, Utc};
+use gateway_core::{engine::execution::ClientAuthenticationError, policy::ClientApiKeyId};
 
 use super::{MutationActor, MutationContext, Revision};
 
@@ -40,43 +41,82 @@ impl AdminRequestContext {
     }
 }
 
-/// 管理员登录命令。
+/// 登录方式只决定凭据验证流程，不能直接授予会话权限。
 #[derive(Clone, PartialEq, Eq)]
-pub struct LoginCommand {
-    pub username: Option<String>,
-    pub password: String,
+pub enum LoginCommand {
+    Admin {
+        username: Option<String>,
+        password: String,
+    },
+    Key {
+        api_key: String,
+    },
 }
 
 impl std::fmt::Debug for LoginCommand {
     fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        formatter
-            .debug_struct("LoginCommand")
-            .field("username", &self.username)
-            .field("password", &"[REDACTED]")
-            .finish()
+        match self {
+            Self::Admin { username, .. } => formatter
+                .debug_struct("AdminLogin")
+                .field("username", username)
+                .field("password", &"[REDACTED]")
+                .finish(),
+            Self::Key { .. } => formatter
+                .debug_struct("KeyLogin")
+                .field("api_key", &"[REDACTED]")
+                .finish(),
+        }
     }
 }
 
 /// 登录成功后返回的会话事实。
-#[derive(Debug, Clone, PartialEq, Eq)]
+#[derive(Clone, PartialEq, Eq)]
 pub struct LoginResult {
     pub session_id: String,
-    pub expires_at: DateTime<Utc>,
+    pub session: AuthSession,
+}
+
+impl std::fmt::Debug for LoginResult {
+    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        formatter
+            .debug_struct("LoginResult")
+            .field("session_id", &"[REDACTED]")
+            .field("session", &self.session)
+            .finish()
+    }
 }
 
 /// 登录状态机可被 API 精确映射的失败类型。
 #[derive(Debug, Clone, Copy, PartialEq, Eq, thiserror::Error)]
 pub enum LoginError {
-    #[error("invalid administrator credentials")]
+    #[error("invalid login credentials")]
     InvalidCredentials,
-    #[error("administrator authentication is unavailable")]
+    #[error("too many login attempts")]
+    TooManyAttempts { retry_after_seconds: u64 },
+    #[error("authentication is unavailable")]
     Unavailable,
 }
 
-/// Redis 中可恢复的管理员会话。
+impl From<ClientAuthenticationError> for LoginError {
+    fn from(error: ClientAuthenticationError) -> Self {
+        match error {
+            ClientAuthenticationError::InvalidKey => Self::InvalidCredentials,
+            ClientAuthenticationError::SnapshotUnavailable => Self::Unavailable,
+        }
+    }
+}
+
+/// 服务端已验证的身份绑定，不保存密码或原始 API Key。
 #[derive(Debug, Clone, PartialEq, Eq)]
-pub struct AdminSession {
-    pub admin_user_id: String,
+pub enum SessionSubject {
+    Admin { admin_user_id: String },
+    Key { client_key_id: ClientApiKeyId },
+}
+
+/// 两种登录方式共用的固定有效期会话。
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct AuthSession {
+    pub subject: SessionSubject,
     pub expires_at: DateTime<Utc>,
 }
 

backend/crates/gateway-admin/src/model/key_usage.rs

@@ -0,0 +1,41 @@
+//! 单页 Key 用量的查询合同；身份范围不由调用方提供。
+
+use super::{
+    PageSize,
+    client_keys::ClientKeyRecord,
+    observability::{
+        HealthTimeline, OpsErrorPage, RequestMetricPoint, TimeRange, UsageOverview, UsagePage,
+    },
+};
+
+#[derive(Debug, Clone)]
+pub struct KeyUsageQuery {
+    pub range: TimeRange,
+    pub model: Option<String>,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub enum KeyUsageRecordKind {
+    Success,
+    Error,
+}
+
+#[derive(Debug, Clone)]
+pub struct KeyUsageRecordsQuery {
+    pub usage: KeyUsageQuery,
+    pub kind: KeyUsageRecordKind,
+    pub current_page: u32,
+    pub page_size: PageSize,
+}
+
+pub struct KeyUsageOverview {
+    pub key: ClientKeyRecord,
+    pub overview: UsageOverview,
+    pub trend: Vec<RequestMetricPoint>,
+    pub health_timeline: HealthTimeline,
+}
+
+pub enum KeyUsageRecords {
+    Success(UsagePage),
+    Error(OpsErrorPage),
+}

backend/crates/gateway-admin/src/model/mod.rs

@@ -8,6 +8,7 @@ pub mod auth;
 pub mod backup;
 pub mod client_distribution;
 pub mod client_keys;
+pub mod key_usage;
 pub mod observability;
 pub mod provider_credentials;
 pub mod proxies;
@@ -21,6 +22,7 @@ pub mod system;
 pub enum AdminErrorKind {
     Invalid,
     Unauthorized,
+    Forbidden,
     NotFound,
     Conflict,
     RateLimited,

backend/crates/gateway-admin/src/model/provider_credentials.rs

@@ -99,6 +99,7 @@ impl fmt::Debug for PrepareCredentialImport {
 /// Provider 已验证、可由 Store 原子创建的一份 credential。
 #[derive(Debug, Clone, PartialEq)]
 pub struct PreparedCredentialCreate {
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
     pub outbound_proxy: Option<gateway_core::account::OutboundProxy>,
     pub account_id: ProviderAccountId,
     pub provider_kind: ProviderKind,

backend/crates/gateway-admin/src/model/settings.rs

@@ -23,6 +23,9 @@ pub struct RuntimeSettings {
     pub refresh_concurrency: u32,
     pub max_concurrent_per_account: u32,
     pub request_interval_ms: u64,
+    pub max_waiting_per_key: u32,
+    pub max_waiting_per_account: u32,
+    pub concurrency_wait_timeout_seconds: u32,
     pub rotation_strategy: RotationStrategy,
     pub min_codex_desktop_version: Option<String>,
     pub min_codex_cli_version: Option<String>,
@@ -40,6 +43,9 @@ pub struct ReplaceRuntimeSettings {
     pub refresh_concurrency: u32,
     pub max_concurrent_per_account: u32,
     pub request_interval_ms: u64,
+    pub max_waiting_per_key: u32,
+    pub max_waiting_per_account: u32,
+    pub concurrency_wait_timeout_seconds: u32,
     pub rotation_strategy: RotationStrategy,
     pub min_codex_desktop_version: Option<String>,
     pub min_codex_cli_version: Option<String>,

backend/crates/gateway-admin/src/ports/store.rs

@@ -2,7 +2,7 @@
 //!
 //! 端口按业务资源拆分，方法使用领域模型，不暴露连接池、事务或 Redis client。
 
-use std::sync::Arc;
+use std::{net::IpAddr, sync::Arc, time::Duration};
 
 use async_trait::async_trait;
 use chrono::{DateTime, Utc};
@@ -20,7 +20,7 @@ use crate::model::{
         AccountUpdateResult, AccountUsage, AccountUsageWindowQuery, AccountUsageWindowResult,
         AccountsUpdateResult, BatchUpdateAccounts, DeleteAccounts, UpdateAccount,
     },
-    auth::{AdminAuditEvent, AdminSession},
+    auth::{AdminAuditEvent, AuthSession},
     client_keys::{
         ClientKeyListQuery, ClientKeyPage, ClientKeyRecord, ClientKeySecret, DeleteClientKey,
         NewClientKey, SetClientKeyEnabled, UpdateClientKey,
@@ -195,7 +195,7 @@ pub trait AccountRuntimeStore: Send + Sync {
     ) -> AdminStoreResult<AccountRuntimeSnapshot>;
 }
 
-/// 管理员密码、会话和安全审计。
+/// 控制面凭据、统一会话、登录限流与管理员安全审计。
 #[async_trait]
 pub trait AuthStore: Send + Sync {
     async fn load_password_hash(&self, admin_user_id: &str) -> AdminStoreResult<Option<String>>;
@@ -208,19 +208,38 @@ pub trait AuthStore: Send + Sync {
 
     async fn load_admin_api_key(&self) -> AdminStoreResult<Option<AdminApiKey>>;
 
-    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>>;
+    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>>;
 
-    async fn store_session(&self, session_id: &str, session: &AdminSession)
-    -> AdminStoreResult<()>;
+    async fn store_session(&self, session_id: &str, session: &AuthSession) -> AdminStoreResult<()>;
 
-    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>>;
+    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>>;
+
+    async fn client_key_enabled(
+        &self,
+        id: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool>;
+
+    /// 原子消费来源桶与全局桶的一次登录尝试；被拒绝时返回建议重试间隔。
+    async fn consume_login_attempt(
+        &self,
+        source_ip: IpAddr,
+        source_limit: u32,
+        global_limit: u32,
+        window: Duration,
+    ) -> AdminStoreResult<Option<Duration>>;
 
     async fn append_audit_event(&self, event: AdminAuditEvent) -> AdminStoreResult<()>;
 }
 
-/// Client API Key 管理写入。
+/// Client API Key 资料读取与管理写入。
 #[async_trait]
 pub trait ClientKeyStore: Send + Sync {
+    /// 按已验证的 ID 读取资料，不读取完整明文 Key。
+    async fn get_client_key(
+        &self,
+        id: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<Option<ClientKeyRecord>>;
+
     async fn list_client_keys(&self, query: ClientKeyListQuery) -> AdminStoreResult<ClientKeyPage>;
 
     async fn reveal_client_key(

backend/crates/gateway-admin/src/use_case/accounts.rs

@@ -570,7 +570,7 @@ impl AccountsService for DefaultAccountsService {
             .await
             .map_err(|error| map_store_error(error, "provider accounts"))?;
         for (provider, provider_ids) in providers.values() {
-            if !enabled {
+            if enabled == Some(false) {
                 for account_id in provider_ids {
                     provider.account_unavailable(account_id).await;
                 }

backend/crates/gateway-admin/src/use_case/auth.rs

@@ -1,20 +1,22 @@
-//! 管理员认证状态机。
+//! 控制面登录、会话恢复与身份校验的唯一 owner。
 
-use std::sync::Arc;
+use std::{net::IpAddr, sync::Arc, time::Duration as StdDuration};
 
 use argon2::{Argon2, PasswordHash, PasswordHasher, PasswordVerifier};
 use async_trait::async_trait;
 use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD};
 use chrono::{Duration, Utc};
+use gateway_core::engine::execution::ClientKeyVerifier;
 use rand_core::{OsRng, RngCore as _};
 use subtle::ConstantTimeEq as _;
 use uuid::Uuid;
 
 use crate::{
     model::{
-        AdminError,
+        AdminError, AdminErrorKind,
         auth::{
-            AdminAuditEvent, AdminSession, AuditActorKind, LoginCommand, LoginError, LoginResult,
+            AdminAuditEvent, AuditActorKind, AuthSession, LoginCommand, LoginError, LoginResult,
+            SessionSubject,
         },
     },
     ports::store::AuthStore,
@@ -22,61 +24,108 @@ use crate::{
 
 use super::map_store_error;
 
-/// API 鉴权与管理员登录消费的统一服务。
+/// 所有控制面接口消费同一个会话服务，权限由服务端身份决定。
 #[async_trait]
 pub trait AuthService: Send + Sync {
     async fn ensure_default_admin(&self, password: &str) -> Result<bool, AdminError>;
+    async fn session(&self, session_id: Option<&str>) -> Result<Option<AuthSession>, AdminError>;
     async fn resolve_admin_user_id(
         &self,
         session_id: Option<&str>,
     ) -> Result<Option<String>, AdminError>;
     async fn verify_admin_api_key(&self, key: &str) -> Result<bool, AdminError>;
-    async fn login(&self, command: LoginCommand) -> Result<LoginResult, LoginError>;
-    async fn validate_session(&self, session_id: Option<&str>) -> Result<bool, AdminError>;
+    async fn login(
+        &self,
+        command: LoginCommand,
+        source_ip: IpAddr,
+        previous_session_id: Option<&str>,
+    ) -> Result<LoginResult, LoginError>;
     async fn logout(&self, session_id: &str) -> Result<(), AdminError>;
 }
 
-/// 会话有效期上限(366 天);超出的配置按上限截断,保证 TTL 构造与
-/// `Utc::now() + session_ttl` 永不越界。
 const MAX_SESSION_TTL_MINUTES: i64 = 366 * 24 * 60;
+const LOGIN_WINDOW: StdDuration = StdDuration::from_secs(60);
+const LOGIN_ATTEMPTS_PER_SOURCE: u32 = 10;
+const LOGIN_ATTEMPTS_GLOBAL: u32 = 200;
 
-/// 单管理员认证策略的最终实现。
 pub(crate) struct DefaultAuthService {
     default_admin_user_id: String,
-    session_ttl: Duration,
+    admin_session_ttl: Duration,
+    key_session_ttl: Duration,
     store: Arc<dyn AuthStore>,
+    verifier: Arc<dyn ClientKeyVerifier>,
 }
 
 impl DefaultAuthService {
     #[must_use]
     pub(crate) fn new(
         default_admin_user_id: impl Into<String>,
-        session_ttl_minutes: u64,
+        admin_session_ttl_minutes: u64,
+        key_session_ttl_minutes: u64,
         store: Arc<dyn AuthStore>,
+        verifier: Arc<dyn ClientKeyVerifier>,
     ) -> Self {
-        let minutes = i64::try_from(session_ttl_minutes)
-            .unwrap_or(MAX_SESSION_TTL_MINUTES)
-            .clamp(1, MAX_SESSION_TTL_MINUTES);
         Self {
             default_admin_user_id: default_admin_user_id.into(),
-            session_ttl: Duration::minutes(minutes),
+            admin_session_ttl: session_ttl(admin_session_ttl_minutes),
+            key_session_ttl: session_ttl(key_session_ttl_minutes),
             store,
+            verifier,
         }
     }
 
-    fn auth_audit(&self, action: &str, occurred_at: chrono::DateTime<Utc>) -> AdminAuditEvent {
+    fn auth_audit(&self, action: &str, admin_user_id: &str) -> AdminAuditEvent {
         AdminAuditEvent {
             id: format!("audit_{}", Uuid::now_v7().simple()),
             actor_kind: AuditActorKind::AdminSession,
-            actor_admin_user_id: Some(self.default_admin_user_id.clone()),
-            actor_ref: crate::model::auth::admin_session_actor_ref(&self.default_admin_user_id),
+            actor_admin_user_id: Some(admin_user_id.to_owned()),
+            actor_ref: crate::model::auth::admin_session_actor_ref(admin_user_id),
             request_id: None,
             action: action.to_owned(),
             entity_kind: "admin_session".to_owned(),
-            entity_ref: self.default_admin_user_id.clone(),
+            entity_ref: admin_user_id.to_owned(),
             config_revision: None,
             changed_fields: Vec::new(),
-            occurred_at,
+            occurred_at: Utc::now(),
+        }
+    }
+
+    async fn authenticate(&self, command: LoginCommand) -> Result<SessionSubject, LoginError> {
+        match command {
+            LoginCommand::Admin { username, password } => {
+                if username.as_deref().unwrap_or(&self.default_admin_user_id)
+                    != self.default_admin_user_id
+                {
+                    return Err(LoginError::InvalidCredentials);
+                }
+                let hash = self
+                    .store
+                    .load_password_hash(&self.default_admin_user_id)
+                    .await
+                    .map_err(|_| LoginError::Unavailable)?
+                    .ok_or(LoginError::InvalidCredentials)?;
+                if !verify_admin_password(&password, &hash).map_err(|_| LoginError::Unavailable)? {
+                    return Err(LoginError::InvalidCredentials);
+                }
+                Ok(SessionSubject::Admin {
+                    admin_user_id: self.default_admin_user_id.clone(),
+                })
+            }
+            LoginCommand::Key { api_key } => {
+                if api_key.is_empty() || api_key.len() > 4096 {
+                    return Err(LoginError::InvalidCredentials);
+                }
+                let client_key_id = self.verifier.verify_client_key(&api_key)?;
+                if !self
+                    .store
+                    .client_key_enabled(&client_key_id)
+                    .await
+                    .map_err(|_| LoginError::Unavailable)?
+                {
+                    return Err(LoginError::InvalidCredentials);
+                }
+                Ok(SessionSubject::Key { client_key_id })
+            }
         }
     }
 }
@@ -91,22 +140,51 @@ impl AuthService for DefaultAuthService {
             .map_err(|error| map_store_error(error, "administrator"))
     }
 
+    async fn session(&self, session_id: Option<&str>) -> Result<Option<AuthSession>, AdminError> {
+        let Some(session_id) = session_id.filter(|value| !value.is_empty()) else {
+            return Ok(None);
+        };
+        let Some(session) = self
+            .store
+            .load_session(session_id)
+            .await
+            .map_err(|error| map_store_error(error, "authentication session"))?
+        else {
+            return Ok(None);
+        };
+        if session.expires_at <= Utc::now() {
+            let _ = self.store.delete_session(session_id).await;
+            return Ok(None);
+        }
+        if let SessionSubject::Key { client_key_id } = &session.subject
+            && !self
+                .store
+                .client_key_enabled(client_key_id)
+                .await
+                .map_err(|error| map_store_error(error, "session key"))?
+        {
+            let _ = self.store.delete_session(session_id).await;
+            return Ok(None);
+        }
+        Ok(Some(session))
+    }
+
     async fn resolve_admin_user_id(
         &self,
         session_id: Option<&str>,
     ) -> Result<Option<String>, AdminError> {
-        let Some(session_id) = session_id else {
-            return Ok(None);
-        };
-        self.store
-            .load_session(session_id)
-            .await
-            .map(|session| {
-                session
-                    .filter(|session| session.expires_at > Utc::now())
-                    .map(|session| session.admin_user_id)
-            })
-            .map_err(|error| map_store_error(error, "administrator session"))
+        match self
+            .session(session_id)
+            .await?
+            .map(|session| session.subject)
+        {
+            Some(SessionSubject::Admin { admin_user_id }) => Ok(Some(admin_user_id)),
+            Some(SessionSubject::Key { .. }) => Err(AdminError::new(
+                AdminErrorKind::Forbidden,
+                "当前身份无权访问管理接口",
+            )),
+            None => Ok(None),
+        }
     }
 
     async fn verify_admin_api_key(&self, key: &str) -> Result<bool, AdminError> {
@@ -124,69 +202,77 @@ impl AuthService for DefaultAuthService {
         }))
     }
 
-    async fn login(&self, command: LoginCommand) -> Result<LoginResult, LoginError> {
-        if command
-            .username
-            .as_deref()
-            .unwrap_or(&self.default_admin_user_id)
-            != self.default_admin_user_id
-        {
-            return Err(LoginError::InvalidCredentials);
-        }
-        let hash = self
+    async fn login(
+        &self,
+        command: LoginCommand,
+        source_ip: IpAddr,
+        previous_session_id: Option<&str>,
+    ) -> Result<LoginResult, LoginError> {
+        if let Some(retry_after) = self
             .store
-            .load_password_hash(&self.default_admin_user_id)
-            .await
-            .map_err(|_| LoginError::Unavailable)?
-            .ok_or(LoginError::InvalidCredentials)?;
-        if !verify_admin_password(&command.password, &hash).map_err(|_| LoginError::Unavailable)? {
-            return Err(LoginError::InvalidCredentials);
-        }
-
-        let session_id = random_session_token();
-        let expires_at = Utc::now() + self.session_ttl;
-        self.store
-            .store_session(
-                &session_id,
-                &AdminSession {
-                    admin_user_id: self.default_admin_user_id.clone(),
-                    expires_at,
-                },
+            .consume_login_attempt(
+                source_ip,
+                LOGIN_ATTEMPTS_PER_SOURCE,
+                LOGIN_ATTEMPTS_GLOBAL,
+                LOGIN_WINDOW,
             )
             .await
-            .map_err(|_| LoginError::Unavailable)?;
-        if self
-            .store
-            .append_audit_event(self.auth_audit("admin.login", Utc::now()))
+            .map_err(|_| LoginError::Unavailable)?
+        {
+            return Err(LoginError::TooManyAttempts {
+                retry_after_seconds: retry_after.as_secs().max(1),
+            });
+        }
+        let subject = self.authenticate(command).await?;
+        let ttl = match subject {
+            SessionSubject::Admin { .. } => self.admin_session_ttl,
+            SessionSubject::Key { .. } => self.key_session_ttl,
+        };
+        let session = AuthSession {
+            subject,
+            expires_at: Utc::now() + ttl,
+        };
+        let session_id = random_session_token();
+        self.store
+            .store_session(&session_id, &session)
             .await
-            .is_err()
+            .map_err(|_| LoginError::Unavailable)?;
+        if let SessionSubject::Admin { admin_user_id } = &session.subject
+            && self
+                .store
+                .append_audit_event(self.auth_audit("admin.login", admin_user_id))
+                .await
+                .is_err()
+        {
+            let _ = self.store.delete_session(&session_id).await;
+            return Err(LoginError::Unavailable);
+        }
+        // 新身份验证成功后才撤销旧会话；撤销失败时不向浏览器提交新会话。
+        if let Some(previous) = previous_session_id.filter(|value| !value.is_empty())
+            && self.logout(previous).await.is_err()
         {
             let _ = self.store.delete_session(&session_id).await;
             return Err(LoginError::Unavailable);
         }
         Ok(LoginResult {
             session_id,
-            expires_at,
+            session,
         })
     }
 
-    async fn validate_session(&self, session_id: Option<&str>) -> Result<bool, AdminError> {
-        Ok(self.resolve_admin_user_id(session_id).await?.is_some())
-    }
-
     async fn logout(&self, session_id: &str) -> Result<(), AdminError> {
         let session = self
             .store
             .delete_session(session_id)
             .await
-            .map_err(|error| map_store_error(error, "administrator session"))?;
-        if let Some(session) = session {
-            let mut event = self.auth_audit("admin.logout", Utc::now());
-            event.actor_admin_user_id = Some(session.admin_user_id.clone());
-            event.actor_ref = crate::model::auth::admin_session_actor_ref(&session.admin_user_id);
-            event.entity_ref = session.admin_user_id;
+            .map_err(|error| map_store_error(error, "authentication session"))?;
+        if let Some(AuthSession {
+            subject: SessionSubject::Admin { admin_user_id },
+            ..
+        }) = session
+        {
             self.store
-                .append_audit_event(event)
+                .append_audit_event(self.auth_audit("admin.logout", &admin_user_id))
                 .await
                 .map_err(|error| map_store_error(error, "administrator audit"))?;
         }
@@ -194,6 +280,14 @@ impl AuthService for DefaultAuthService {
     }
 }
 
+fn session_ttl(minutes: u64) -> Duration {
+    Duration::minutes(
+        i64::try_from(minutes)
+            .unwrap_or(MAX_SESSION_TTL_MINUTES)
+            .clamp(1, MAX_SESSION_TTL_MINUTES),
+    )
+}
+
 fn hash_admin_password(password: &str) -> Result<String, AdminError> {
     Argon2::default()
         .hash_password(password.as_bytes())

backend/crates/gateway-admin/src/use_case/key_usage.rs

@@ -0,0 +1,165 @@
+//! Key 自助查询只从统一会话取得范围，复用现有观测和额度账本。
+
+use std::sync::Arc;
+
+use async_trait::async_trait;
+use chrono::Utc;
+
+use crate::{
+    AuthService,
+    model::{
+        AdminError, AdminErrorKind,
+        auth::SessionSubject,
+        key_usage::{
+            KeyUsageOverview, KeyUsageQuery, KeyUsageRecordKind, KeyUsageRecords,
+            KeyUsageRecordsQuery,
+        },
+        observability::{
+            OpsErrorFilter, OpsErrorQuery, TimeRange, UsageFilter, UsageQuery, china_day_start,
+        },
+    },
+    ports::store::{ClientKeyStore, ObservabilityStore},
+};
+use gateway_core::policy::ClientApiKeyId;
+
+use super::{map_store_error, observability::health_timeline_at};
+
+#[async_trait]
+pub trait KeyUsageService: Send + Sync {
+    async fn overview(
+        &self,
+        session_id: Option<&str>,
+        query: KeyUsageQuery,
+    ) -> Result<Option<KeyUsageOverview>, AdminError>;
+
+    async fn records(
+        &self,
+        session_id: Option<&str>,
+        query: KeyUsageRecordsQuery,
+    ) -> Result<Option<KeyUsageRecords>, AdminError>;
+}
+
+pub(crate) struct DefaultKeyUsageService {
+    auth: Arc<dyn AuthService>,
+    keys: Arc<dyn ClientKeyStore>,
+    observations: Arc<dyn ObservabilityStore>,
+}
+
+impl DefaultKeyUsageService {
+    pub(crate) fn new(
+        auth: Arc<dyn AuthService>,
+        keys: Arc<dyn ClientKeyStore>,
+        observations: Arc<dyn ObservabilityStore>,
+    ) -> Self {
+        Self {
+            auth,
+            keys,
+            observations,
+        }
+    }
+
+    async fn key_id(&self, session_id: Option<&str>) -> Result<Option<ClientApiKeyId>, AdminError> {
+        match self
+            .auth
+            .session(session_id)
+            .await?
+            .map(|session| session.subject)
+        {
+            Some(SessionSubject::Key { client_key_id }) => Ok(Some(client_key_id)),
+            Some(SessionSubject::Admin { .. }) => Err(AdminError::new(
+                AdminErrorKind::Forbidden,
+                "当前身份无权访问密钥用量接口",
+            )),
+            None => Ok(None),
+        }
+    }
+}
+
+fn usage_filter(id: &ClientApiKeyId, model: Option<String>) -> UsageFilter {
+    UsageFilter {
+        client_api_key_ref: Some(id.as_str().to_owned()),
+        model,
+        ..UsageFilter::default()
+    }
+}
+
+#[async_trait]
+impl KeyUsageService for DefaultKeyUsageService {
+    async fn overview(
+        &self,
+        session_id: Option<&str>,
+        query: KeyUsageQuery,
+    ) -> Result<Option<KeyUsageOverview>, AdminError> {
+        let Some(id) = self.key_id(session_id).await? else {
+            return Ok(None);
+        };
+        let Some(key) = self
+            .keys
+            .get_client_key(&id)
+            .await
+            .map_err(|error| map_store_error(error, "key usage profile"))?
+            .filter(|key| key.enabled)
+        else {
+            return Ok(None);
+        };
+        let filter = usage_filter(&id, query.model);
+        let now = Utc::now();
+        // 健康条始终展示北京时间今日，不随历史范围或模型筛选改变。
+        let today = TimeRange {
+            start: china_day_start(now),
+            end: now,
+        };
+        let (overview, trend, health_points) = futures::try_join!(
+            self.observations.usage_summary(query.range, filter.clone()),
+            self.observations.usage_trend(query.range, filter),
+            self.observations
+                .usage_trend(today, usage_filter(&id, None)),
+        )
+        .map_err(|error| map_store_error(error, "key usage overview"))?;
+        Ok(Some(KeyUsageOverview {
+            key,
+            overview,
+            trend,
+            health_timeline: health_timeline_at(&health_points, now),
+        }))
+    }
+
+    async fn records(
+        &self,
+        session_id: Option<&str>,
+        query: KeyUsageRecordsQuery,
+    ) -> Result<Option<KeyUsageRecords>, AdminError> {
+        let Some(id) = self.key_id(session_id).await? else {
+            return Ok(None);
+        };
+        let result = match query.kind {
+            KeyUsageRecordKind::Success => KeyUsageRecords::Success(
+                self.observations
+                    .list_usage_records(UsageQuery {
+                        range: query.usage.range,
+                        filter: usage_filter(&id, query.usage.model),
+                        current_page: query.current_page,
+                        page_size: query.page_size,
+                    })
+                    .await
+                    .map_err(|error| map_store_error(error, "key usage records"))?,
+            ),
+            KeyUsageRecordKind::Error => KeyUsageRecords::Error(
+                self.observations
+                    .list_ops_errors(OpsErrorQuery {
+                        range: query.usage.range,
+                        filter: OpsErrorFilter {
+                            client_api_key_ref: Some(id.as_str().to_owned()),
+                            model: query.usage.model,
+                            ..OpsErrorFilter::default()
+                        },
+                        current_page: query.current_page,
+                        page_size: query.page_size,
+                    })
+                    .await
+                    .map_err(|error| map_store_error(error, "key usage errors"))?,
+            ),
+        };
+        Ok(Some(result))
+    }
+}

backend/crates/gateway-admin/src/use_case/mod.rs

@@ -6,6 +6,7 @@ pub mod auth;
 pub mod backup;
 pub mod client_distribution;
 pub mod client_keys;
+pub mod key_usage;
 pub mod observability;
 pub mod openai;
 pub mod proxies;

backend/crates/gateway-admin/src/use_case/observability.rs

@@ -852,7 +852,10 @@ impl DefaultObservabilityService {
 
 /// 按指定时刻计算中国自然日的 96 个 15 分钟健康桶。
 #[must_use]
-fn health_timeline_at(records: &[RequestMetricPoint], now: DateTime<Utc>) -> HealthTimeline {
+pub(super) fn health_timeline_at(
+    records: &[RequestMetricPoint],
+    now: DateTime<Utc>,
+) -> HealthTimeline {
     let current_slot = quarter_hour_start(now);
     let start = china_day_start(now);
     let mut buckets = (0..HEALTH_TIMELINE_SLOTS)

backend/crates/gateway-admin/src/use_case/settings.rs

@@ -117,6 +117,9 @@ fn validate_settings(command: &ReplaceRuntimeSettings) -> Result<(), AdminError>
     let valid = command.refresh_margin_seconds > 0
         && command.refresh_concurrency > 0
         && command.max_concurrent_per_account > 0
+        && command.max_waiting_per_key <= 1_000
+        && command.max_waiting_per_account <= 1_000
+        && (1..=120).contains(&command.concurrency_wait_timeout_seconds)
         && command.usage_retention_days >= 31
         && command.ops_event_retention_days > 0
         && command.audit_retention_days > 0

backend/crates/gateway-admin/tests/backup/support.rs

@@ -633,22 +633,39 @@ impl AuthStore for FakeAuthStore {
     async fn load_session(
         &self,
         _session_id: &str,
-    ) -> AdminStoreResult<Option<gateway_admin::model::auth::AdminSession>> {
+    ) -> AdminStoreResult<Option<gateway_admin::model::auth::AuthSession>> {
         Ok(None)
     }
     async fn store_session(
         &self,
         _session_id: &str,
-        _session: &gateway_admin::model::auth::AdminSession,
+        _session: &gateway_admin::model::auth::AuthSession,
     ) -> AdminStoreResult<()> {
         Ok(())
     }
     async fn delete_session(
         &self,
         _session_id: &str,
-    ) -> AdminStoreResult<Option<gateway_admin::model::auth::AdminSession>> {
+    ) -> AdminStoreResult<Option<gateway_admin::model::auth::AuthSession>> {
         Ok(None)
     }
+    async fn client_key_enabled(
+        &self,
+        _: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool> {
+        Ok(false)
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        _: std::net::IpAddr,
+        _: u32,
+        _: u32,
+        _: std::time::Duration,
+    ) -> AdminStoreResult<Option<std::time::Duration>> {
+        Ok(None)
+    }
+
     async fn append_audit_event(&self, event: AdminAuditEvent) -> AdminStoreResult<()> {
         self.audit.lock().expect("audit").push(event);
         Ok(())

backend/crates/gateway-admin/tests/main.rs

@@ -23,6 +23,14 @@ mod backup;
 mod model;
 mod use_case;
 
+#[test]
+fn client_config_should_default_to_a_one_day_session() {
+    assert_eq!(
+        gateway_admin::ClientConfig::default().session_ttl_minutes,
+        1440
+    );
+}
+
 #[test]
 fn revision_should_reject_zero() {
     assert!(Revision::new(0).is_err());
@@ -88,7 +96,7 @@ fn admin_api_key_debug_should_redact_plaintext() {
 
 #[test]
 fn login_command_debug_should_redact_password() {
-    let command = LoginCommand {
+    let command = LoginCommand::Admin {
         username: None,
         password: "secret-password".to_owned(),
     };

backend/crates/gateway-admin/tests/use_case/accounts.rs

@@ -922,6 +922,9 @@ impl SettingsStore for StaticSettingsStore {
             refresh_concurrency: 2,
             max_concurrent_per_account: 1,
             request_interval_ms: 0,
+            max_waiting_per_key: 0,
+            max_waiting_per_account: 0,
+            concurrency_wait_timeout_seconds: 30,
             rotation_strategy: RotationStrategy::Smart,
             min_codex_desktop_version: None,
             min_codex_cli_version: None,
@@ -1330,6 +1333,8 @@ async fn accounts_update_should_commit_then_release_disabled_account_and_publish
         .update(
             &context("update-request"),
             UpdateAccount {
+                notes: None,
+                model_access: Default::default(),
                 outbound_proxy: None,
                 account_id: "acct_test".to_owned(),
                 enabled: false,
@@ -1367,6 +1372,8 @@ async fn accounts_update_should_not_notify_provider_when_store_commit_fails() {
         .update(
             &context("update-failure"),
             UpdateAccount {
+                notes: None,
+                model_access: Default::default(),
                 outbound_proxy: None,
                 account_id: "acct_test".to_owned(),
                 enabled: false,
@@ -1409,12 +1416,13 @@ async fn accounts_batch_update_should_commit_once_and_notify_each_provider() {
         .batch_update(
             &context("batch-update-request"),
             BatchUpdateAccounts {
+                model_access: Default::default(),
                 outbound_proxy: None,
                 account_ids: vec!["acct_openai".to_owned(), "acct_xai".to_owned()],
-                enabled: false,
-                concurrency_limit: None,
-                weight: gateway_core::account::AccountWeight::DEFAULT,
-                group_ids: Vec::new(),
+                enabled: Some(false),
+                concurrency_limit: Some(None),
+                weight: Some(gateway_core::account::AccountWeight::DEFAULT),
+                group_ids: Some(Vec::new()),
             },
         )
         .await
@@ -2400,6 +2408,8 @@ fn quota_local_usage(account_id: &str, total_tokens: u64) -> AccountUsage {
 pub(super) fn account_record(kind: &str) -> AccountRecord {
     let now = Utc::now();
     AccountRecord {
+        notes: None,
+        model_access: Default::default(),
         outbound_proxy: None,
         id: "acct_test".to_owned(),
         provider_kind: ProviderKind::new(kind).expect("provider kind"),
@@ -2442,6 +2452,7 @@ fn prepared_create_with_id(
 ) -> PreparedCredentialCreate {
     let now = Utc::now();
     PreparedCredentialCreate {
+        model_access: Default::default(),
         outbound_proxy: None,
         account_id: ProviderAccountId::new(account_id).expect("prepared account ID"),
         provider_kind,
@@ -2733,6 +2744,8 @@ fn unsupported() -> ProviderAdminError {
 
 pub(super) fn import_settings() -> gateway_admin::model::accounts::AccountImportSettings {
     gateway_admin::model::accounts::AccountImportSettings {
+        notes: Some("团队备用".to_owned()),
+        model_access: Default::default(),
         enabled: false,
         concurrency_limit: Some(
             gateway_core::account::AccountConcurrencyLimit::new(3).expect("concurrency"),

backend/crates/gateway-admin/tests/use_case/auth.rs

@@ -5,7 +5,7 @@ use chrono::{TimeDelta, Utc};
 
 use gateway_admin::{
     model::{
-        auth::{AdminAuditEvent, AdminSession, LoginCommand},
+        auth::{AdminAuditEvent, AuthSession, LoginCommand},
         settings::AdminApiKey,
     },
     ports::store::{AdminStoreResult, AuthStore},
@@ -13,8 +13,11 @@ use gateway_admin::{
 
 #[derive(Default)]
 struct MemoryAuthStore {
+    retry_after: Mutex<Option<std::time::Duration>>,
+    unavailable: std::sync::atomic::AtomicBool,
+    reject_delete: Mutex<Option<String>>,
     password_hash: Mutex<Option<String>>,
-    sessions: Mutex<BTreeMap<String, AdminSession>>,
+    sessions: Mutex<BTreeMap<String, AuthSession>>,
     audits: Mutex<Vec<AdminAuditEvent>>,
 }
 
@@ -41,7 +44,10 @@ impl AuthStore for MemoryAuthStore {
         Ok(None)
     }
 
-    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        if self.unavailable.load(std::sync::atomic::Ordering::SeqCst) {
+            return Err(super::unavailable("session"));
+        }
         Ok(self
             .sessions
             .lock()
@@ -50,11 +56,7 @@ impl AuthStore for MemoryAuthStore {
             .cloned())
     }
 
-    async fn store_session(
-        &self,
-        session_id: &str,
-        session: &AdminSession,
-    ) -> AdminStoreResult<()> {
+    async fn store_session(&self, session_id: &str, session: &AuthSession) -> AdminStoreResult<()> {
         self.sessions
             .lock()
             .expect("sessions")
@@ -62,10 +64,30 @@ impl AuthStore for MemoryAuthStore {
         Ok(())
     }
 
-    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        if self.reject_delete.lock().unwrap().as_deref() == Some(session_id) {
+            return Err(super::unavailable("delete session"));
+        }
         Ok(self.sessions.lock().expect("sessions").remove(session_id))
     }
 
+    async fn client_key_enabled(
+        &self,
+        _: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool> {
+        Ok(false)
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        _: std::net::IpAddr,
+        _: u32,
+        _: u32,
+        _: std::time::Duration,
+    ) -> AdminStoreResult<Option<std::time::Duration>> {
+        Ok(*self.retry_after.lock().unwrap())
+    }
+
     async fn append_audit_event(&self, event: AdminAuditEvent) -> AdminStoreResult<()> {
         self.audits.lock().expect("audits").push(event);
         Ok(())
@@ -79,19 +101,24 @@ async fn successful_login_should_create_expiring_session_and_audit() {
 
     let result = services
         .auth()
-        .login(LoginCommand {
-            username: Some("admin".to_owned()),
-            password: "strong-test-password".to_owned(),
-        })
+        .login(
+            LoginCommand::Admin {
+                username: Some("admin".to_owned()),
+                password: "strong-test-password".to_owned(),
+            },
+            std::net::Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
         .await
         .expect("login");
 
     assert!(
         services
             .auth()
-            .validate_session(Some(&result.session_id))
+            .session(Some(&result.session_id))
             .await
             .expect("validate")
+            .is_some()
     );
     assert_eq!(store.audits.lock().expect("audits").len(), 1);
 }
@@ -113,10 +140,14 @@ async fn repeated_default_initialization_should_not_replace_password() {
     assert!(
         services
             .auth()
-            .login(LoginCommand {
-                username: Some("admin".to_owned()),
-                password: "first-strong-password".to_owned(),
-            })
+            .login(
+                LoginCommand::Admin {
+                    username: Some("admin".to_owned()),
+                    password: "first-strong-password".to_owned(),
+                },
+                std::net::Ipv4Addr::LOCALHOST.into(),
+                None
+            )
             .await
             .is_ok()
     );
@@ -134,20 +165,134 @@ async fn login_with_huge_session_ttl_should_clamp_expiry_instead_of_panicking()
     let before = Utc::now();
     let result = services
         .auth()
-        .login(LoginCommand {
-            username: Some("admin".to_owned()),
-            password: "strong-test-password".to_owned(),
-        })
+        .login(
+            LoginCommand::Admin {
+                username: Some("admin".to_owned()),
+                password: "strong-test-password".to_owned(),
+            },
+            std::net::Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
         .await
         .expect("login with huge session TTL");
 
-    assert!(result.expires_at > before);
-    assert!(result.expires_at <= before + TimeDelta::days(366) + TimeDelta::minutes(1));
+    assert!(result.session.expires_at > before);
+    assert!(result.session.expires_at <= before + TimeDelta::days(366) + TimeDelta::minutes(1));
     assert!(
         services
             .auth()
-            .validate_session(Some(&result.session_id))
+            .session(Some(&result.session_id))
             .await
             .expect("validate")
+            .is_some()
     );
 }
+
+#[tokio::test]
+async fn expired_sessions_are_removed_and_store_outages_are_not_treated_as_logout() {
+    use gateway_admin::model::{AdminErrorKind, auth::SessionSubject};
+    let store = std::sync::Arc::new(MemoryAuthStore::default());
+    let services = super::AdminHarness::new().auth(store.clone()).build().await;
+    store.sessions.lock().unwrap().insert(
+        "expired".to_owned(),
+        AuthSession {
+            subject: SessionSubject::Admin {
+                admin_user_id: "admin".to_owned(),
+            },
+            expires_at: Utc::now() - TimeDelta::seconds(1),
+        },
+    );
+    assert!(
+        services
+            .auth()
+            .session(Some("expired"))
+            .await
+            .unwrap()
+            .is_none()
+    );
+    assert!(store.sessions.lock().unwrap().is_empty());
+    store
+        .unavailable
+        .store(true, std::sync::atomic::Ordering::SeqCst);
+    assert_eq!(
+        services
+            .auth()
+            .session(Some("existing"))
+            .await
+            .unwrap_err()
+            .kind(),
+        AdminErrorKind::Unavailable
+    );
+}
+
+#[tokio::test]
+async fn both_login_types_share_the_login_limit_before_credentials_are_checked() {
+    use gateway_admin::model::auth::LoginError;
+    let store = std::sync::Arc::new(MemoryAuthStore::default());
+    *store.retry_after.lock().unwrap() = Some(std::time::Duration::from_secs(37));
+    let services = super::AdminHarness::new().auth(store).build().await;
+    for command in [
+        LoginCommand::Admin {
+            username: None,
+            password: "wrong-password".to_owned(),
+        },
+        LoginCommand::Key {
+            api_key: "unknown-key".to_owned(),
+        },
+    ] {
+        assert_eq!(
+            services
+                .auth()
+                .login(command, std::net::Ipv4Addr::LOCALHOST.into(), None)
+                .await
+                .unwrap_err(),
+            LoginError::TooManyAttempts {
+                retry_after_seconds: 37
+            }
+        );
+    }
+}
+
+#[tokio::test]
+async fn failed_rotation_discards_the_new_session_and_leaves_the_old_session_retryable() {
+    use gateway_admin::model::auth::LoginError;
+    let store = std::sync::Arc::new(MemoryAuthStore::default());
+    let services = super::AdminHarness::new().auth(store.clone()).build().await;
+    let command = || LoginCommand::Admin {
+        username: None,
+        password: "strong-test-password".to_owned(),
+    };
+    let original = services
+        .auth()
+        .login(command(), std::net::Ipv4Addr::LOCALHOST.into(), None)
+        .await
+        .unwrap();
+    assert!(!format!("{original:?}").contains(&original.session_id));
+    *store.reject_delete.lock().unwrap() = Some(original.session_id.clone());
+    assert_eq!(
+        services
+            .auth()
+            .login(
+                command(),
+                std::net::Ipv4Addr::LOCALHOST.into(),
+                Some(&original.session_id)
+            )
+            .await
+            .unwrap_err(),
+        LoginError::Unavailable
+    );
+    assert_eq!(store.sessions.lock().unwrap().len(), 1);
+    assert!(
+        services
+            .auth()
+            .session(Some(&original.session_id))
+            .await
+            .unwrap()
+            .is_some()
+    );
+    assert!(services.auth().logout(&original.session_id).await.is_err());
+    *store.reject_delete.lock().unwrap() = None;
+    services.auth().logout(&original.session_id).await.unwrap();
+    services.auth().logout(&original.session_id).await.unwrap();
+    assert!(store.sessions.lock().unwrap().is_empty());
+}

backend/crates/gateway-admin/tests/use_case/auth_key.rs

@@ -0,0 +1,293 @@
+use std::{
+    collections::HashMap,
+    net::{IpAddr, Ipv4Addr},
+    sync::{
+        Arc, Mutex,
+        atomic::{AtomicUsize, Ordering},
+    },
+    time::Duration as StdDuration,
+};
+
+use async_trait::async_trait;
+use chrono::{Duration, Utc};
+use gateway_admin::{
+    model::auth::{AdminAuditEvent, AuthSession, LoginCommand, LoginError, SessionSubject},
+    ports::store::{AdminStoreResult, AuthStore},
+};
+use gateway_core::{
+    engine::execution::{ClientAuthenticationError, ClientKeyVerifier},
+    policy::ClientApiKeyId,
+};
+
+#[test]
+fn client_login_command_debug_should_redact_api_key() {
+    let command = LoginCommand::Key {
+        api_key: "cpr_should_never_be_logged".to_owned(),
+    };
+
+    let rendered = format!("{command:?}");
+
+    assert!(rendered.contains("[REDACTED]"));
+    assert!(!rendered.contains("cpr_should_never_be_logged"));
+}
+
+#[tokio::test]
+async fn login_should_create_key_bound_session_with_fixed_ttl() {
+    let key = client_key("key-42");
+    let store = Arc::new(MemoryKeyAuthStore::new(Some(key.clone())));
+    let verifier = Arc::new(FixtureVerifier::accepting(key.clone()));
+    let services = super::AdminHarness::new()
+        .auth(store.clone())
+        .client_key_verifier(verifier.clone())
+        .client_session_ttl_minutes(30)
+        .build()
+        .await;
+
+    let login = services
+        .auth()
+        .login(
+            login_command("valid-client-key"),
+            Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
+        .await
+        .expect("client login");
+
+    assert!(login.session_id.starts_with("session_"));
+    assert!(!login.session_id.contains("valid-client-key"));
+    assert_eq!(verifier.calls(), 1);
+    assert_eq!(
+        store.session(&login.session_id).map(|value| value.subject),
+        Some(SessionSubject::Key {
+            client_key_id: key.clone()
+        })
+    );
+    let remaining = login.session.expires_at - Utc::now();
+    assert!(remaining > Duration::minutes(29));
+    assert!(remaining <= Duration::minutes(30));
+}
+
+#[tokio::test]
+async fn invalid_and_removed_keys_should_share_the_same_login_error() {
+    let rejected_store = Arc::new(MemoryKeyAuthStore::new(Some(client_key("key-42"))));
+    let rejected = super::AdminHarness::new()
+        .auth(rejected_store.clone())
+        .client_key_verifier(Arc::new(FixtureVerifier::rejecting()))
+        .build()
+        .await
+        .auth()
+        .login(
+            login_command("unknown-key"),
+            Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
+        .await;
+
+    let removed = super::AdminHarness::new()
+        .auth(Arc::new(MemoryKeyAuthStore::new(None)))
+        .client_key_verifier(Arc::new(FixtureVerifier::accepting(
+            ClientApiKeyId::new("removed-key").expect("key id"),
+        )))
+        .build()
+        .await
+        .auth()
+        .login(
+            login_command("disabled-key"),
+            Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
+        .await;
+
+    assert_eq!(rejected, Err(LoginError::InvalidCredentials));
+    assert_eq!(removed, Err(LoginError::InvalidCredentials));
+}
+
+#[tokio::test]
+async fn rate_limit_should_reject_before_verifying_the_candidate_key() {
+    let store = Arc::new(MemoryKeyAuthStore::new(Some(client_key("key-42"))));
+    store.set_retry_after(Some(StdDuration::from_secs(37)));
+    let verifier = Arc::new(FixtureVerifier::accepting(
+        ClientApiKeyId::new("key-42").expect("key id"),
+    ));
+    let services = super::AdminHarness::new()
+        .auth(store.clone())
+        .client_key_verifier(verifier.clone())
+        .build()
+        .await;
+
+    let result = services
+        .auth()
+        .login(
+            login_command("valid-client-key"),
+            Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
+        .await;
+
+    assert_eq!(
+        result,
+        Err(LoginError::TooManyAttempts {
+            retry_after_seconds: 37,
+        })
+    );
+    assert_eq!(verifier.calls(), 0);
+}
+
+#[tokio::test]
+async fn restored_session_should_be_revoked_when_its_key_is_no_longer_active() {
+    let key = client_key("key-42");
+    let store = Arc::new(MemoryKeyAuthStore::new(Some(key.clone())));
+    let services = super::AdminHarness::new()
+        .auth(store.clone())
+        .client_key_verifier(Arc::new(FixtureVerifier::accepting(key.clone())))
+        .build()
+        .await;
+    let login = services
+        .auth()
+        .login(
+            login_command("valid-client-key"),
+            Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
+        .await
+        .expect("client login");
+    store.replace_key(None);
+
+    let session = services
+        .auth()
+        .session(Some(&login.session_id))
+        .await
+        .expect("session status");
+
+    assert!(session.is_none());
+    assert!(store.session(&login.session_id).is_none());
+}
+
+fn login_command(api_key: &str) -> LoginCommand {
+    LoginCommand::Key {
+        api_key: api_key.to_owned(),
+    }
+}
+
+fn client_key(id: &str) -> ClientApiKeyId {
+    ClientApiKeyId::new(id).expect("client key id")
+}
+
+struct FixtureVerifier {
+    result: Result<ClientApiKeyId, ClientAuthenticationError>,
+    calls: AtomicUsize,
+}
+
+impl FixtureVerifier {
+    fn accepting(id: ClientApiKeyId) -> Self {
+        Self {
+            result: Ok(id),
+            calls: AtomicUsize::new(0),
+        }
+    }
+
+    fn rejecting() -> Self {
+        Self {
+            result: Err(ClientAuthenticationError::InvalidKey),
+            calls: AtomicUsize::new(0),
+        }
+    }
+
+    fn calls(&self) -> usize {
+        self.calls.load(Ordering::Relaxed)
+    }
+}
+
+impl ClientKeyVerifier for FixtureVerifier {
+    fn verify_client_key(&self, _: &str) -> Result<ClientApiKeyId, ClientAuthenticationError> {
+        self.calls.fetch_add(1, Ordering::Relaxed);
+        self.result.clone()
+    }
+}
+
+struct MemoryKeyAuthStore {
+    key: Mutex<Option<ClientApiKeyId>>,
+    sessions: Mutex<HashMap<String, AuthSession>>,
+    retry_after: Mutex<Option<StdDuration>>,
+}
+
+impl MemoryKeyAuthStore {
+    fn new(key: Option<ClientApiKeyId>) -> Self {
+        Self {
+            key: Mutex::new(key),
+            sessions: Mutex::new(HashMap::new()),
+            retry_after: Mutex::new(None),
+        }
+    }
+
+    fn replace_key(&self, key: Option<ClientApiKeyId>) {
+        *self.key.lock().expect("client key") = key;
+    }
+
+    fn set_retry_after(&self, retry_after: Option<StdDuration>) {
+        *self.retry_after.lock().expect("retry after") = retry_after;
+    }
+
+    fn session(&self, session_id: &str) -> Option<AuthSession> {
+        self.sessions
+            .lock()
+            .expect("client sessions")
+            .get(session_id)
+            .cloned()
+    }
+}
+
+#[async_trait]
+impl AuthStore for MemoryKeyAuthStore {
+    async fn load_password_hash(&self, _: &str) -> AdminStoreResult<Option<String>> {
+        Ok(None)
+    }
+    async fn create_password_hash_if_absent(&self, _: &str, _: &str) -> AdminStoreResult<bool> {
+        Ok(true)
+    }
+    async fn load_admin_api_key(
+        &self,
+    ) -> AdminStoreResult<Option<gateway_admin::model::settings::AdminApiKey>> {
+        Ok(None)
+    }
+    async fn append_audit_event(&self, _: AdminAuditEvent) -> AdminStoreResult<()> {
+        Ok(())
+    }
+    async fn client_key_enabled(&self, id: &ClientApiKeyId) -> AdminStoreResult<bool> {
+        Ok(self
+            .key
+            .lock()
+            .expect("client key")
+            .as_ref()
+            .is_some_and(|key| key == id))
+    }
+    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        Ok(self.session(session_id))
+    }
+
+    async fn store_session(&self, session_id: &str, session: &AuthSession) -> AdminStoreResult<()> {
+        self.sessions
+            .lock()
+            .expect("client sessions")
+            .insert(session_id.to_owned(), session.clone());
+        Ok(())
+    }
+
+    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        Ok(self
+            .sessions
+            .lock()
+            .expect("client sessions")
+            .remove(session_id))
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        _: IpAddr,
+        _: u32,
+        _: u32,
+        _: StdDuration,
+    ) -> AdminStoreResult<Option<StdDuration>> {
+        Ok(*self.retry_after.lock().expect("retry after"))
+    }
+}

backend/crates/gateway-admin/tests/use_case/client_keys.rs

@@ -26,6 +26,13 @@ struct TestClientKeyStore {
 
 #[async_trait]
 impl ClientKeyStore for TestClientKeyStore {
+    async fn get_client_key(
+        &self,
+        _: &ClientApiKeyId,
+    ) -> AdminStoreResult<Option<ClientKeyRecord>> {
+        Err(unused())
+    }
+
     async fn list_client_keys(&self, query: ClientKeyListQuery) -> AdminStoreResult<ClientKeyPage> {
         assert_eq!(query.page_size.get(), u16::MAX);
         Ok(ClientKeyPage {

backend/crates/gateway-admin/tests/use_case/mod.rs

@@ -1,6 +1,7 @@
 mod account_groups;
 mod accounts;
 mod auth;
+mod auth_key;
 mod backup;
 mod client_keys;
 mod observability;
@@ -19,7 +20,7 @@ use async_trait::async_trait;
 use chrono::{DateTime, Utc};
 use futures::future::BoxFuture;
 use gateway_admin::{
-    AdminConfig, AdminServices, InitialAdminPassword,
+    AdminConfig, AdminServices, ClientConfig, InitialAdminPassword,
     model::{
         MutationContext, Revision,
         account_groups::{
@@ -31,7 +32,7 @@ use gateway_admin::{
             AccountUsage, AccountUsageWindowQuery, AccountUsageWindowResult, AccountsUpdateResult,
             BatchUpdateAccounts, DeleteAccounts, UpdateAccount,
         },
-        auth::{AdminAuditEvent, AdminSession},
+        auth::{AdminAuditEvent, AuthSession},
         client_distribution::CodexDesktopWindowsDownloads,
         client_keys::{
             ClientKeyListQuery, ClientKeyPage, ClientKeyRecord, ClientKeySecret, DeleteClientKey,
@@ -71,7 +72,10 @@ use gateway_admin::{
 };
 use gateway_core::{
     account::ProviderAccountId,
-    engine::probe::{AccountProbe, AccountProbeError, AccountProbeRequest, AccountProbeResult},
+    engine::{
+        execution::{ClientAuthenticationError, ClientKeyVerifier},
+        probe::{AccountProbe, AccountProbeError, AccountProbeRequest, AccountProbeResult},
+    },
     error::{GatewayError, GatewayErrorKind},
     policy::ClientApiKeyId,
     routing::{ConfigRevision, ProviderKind},
@@ -81,6 +85,7 @@ use gateway_core::{
 pub(super) struct AdminHarness {
     default_password: String,
     session_ttl_minutes: u64,
+    client_session_ttl_minutes: u64,
     accounts: Arc<dyn AccountStore>,
     proxies: Arc<dyn gateway_admin::ports::proxy::ProxyStore>,
     account_runtime: Arc<dyn AccountRuntimeStore>,
@@ -93,6 +98,7 @@ pub(super) struct AdminHarness {
     providers: Vec<Arc<dyn ProviderAdmin>>,
     probe: Arc<dyn AccountProbe>,
     system: Arc<dyn SystemOperations>,
+    client_key_verifier: Arc<dyn ClientKeyVerifier>,
 }
 
 impl AdminHarness {
@@ -101,6 +107,7 @@ impl AdminHarness {
         Self {
             default_password: "strong-test-password".to_owned(),
             session_ttl_minutes: 60,
+            client_session_ttl_minutes: 1_440,
             accounts: unavailable.clone(),
             proxies: Arc::new(proxies::TestProxies::default()),
             account_runtime: unavailable.clone(),
@@ -116,6 +123,7 @@ impl AdminHarness {
             ],
             probe: Arc::new(UnavailableProbe),
             system: Arc::new(UnavailableSystem),
+            client_key_verifier: Arc::new(UnavailableClientKeyVerifier),
         }
     }
 
@@ -154,6 +162,16 @@ impl AdminHarness {
         self
     }
 
+    pub(super) fn client_session_ttl_minutes(mut self, minutes: u64) -> Self {
+        self.client_session_ttl_minutes = minutes;
+        self
+    }
+
+    pub(super) fn client_key_verifier(mut self, verifier: Arc<dyn ClientKeyVerifier>) -> Self {
+        self.client_key_verifier = verifier;
+        self
+    }
+
     pub(super) fn observability(mut self, store: Arc<dyn ObservabilityStore>) -> Self {
         self.observability = store;
         self
@@ -201,6 +219,9 @@ impl AdminHarness {
                 default_username: "admin".to_owned(),
                 default_password: InitialAdminPassword::new(self.default_password),
             },
+            ClientConfig {
+                session_ttl_minutes: self.client_session_ttl_minutes,
+            },
             AdminStorePorts::new(
                 AdminAccountStorePorts::new(
                     self.accounts,
@@ -214,11 +235,15 @@ impl AdminHarness {
                 self.settings,
                 self.backup,
             ),
-            self.providers,
-            Arc::new(NoopSnapshot),
-            (self.probe, Arc::new(proxies::TestProxies::default())),
-            Arc::new(NoopClientDistribution),
-            self.system,
+            gateway_admin::AdminRuntimePorts {
+                providers: self.providers,
+                snapshot: Arc::new(NoopSnapshot),
+                account_probe: self.probe,
+                proxy_probe: Arc::new(proxies::TestProxies::default()),
+                client_distribution: Arc::new(NoopClientDistribution),
+                system: self.system,
+                client_key_verifier: self.client_key_verifier,
+            },
         )
         .await
         .expect("initialize admin test harness")
@@ -268,18 +293,35 @@ impl AuthStore for BootstrapAuthStore {
         Ok(None)
     }
 
-    async fn load_session(&self, _: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn load_session(&self, _: &str) -> AdminStoreResult<Option<AuthSession>> {
         Ok(None)
     }
 
-    async fn store_session(&self, _: &str, _: &AdminSession) -> AdminStoreResult<()> {
+    async fn store_session(&self, _: &str, _: &AuthSession) -> AdminStoreResult<()> {
         Err(unavailable("admin session"))
     }
 
-    async fn delete_session(&self, _: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn delete_session(&self, _: &str) -> AdminStoreResult<Option<AuthSession>> {
         Err(unavailable("admin session"))
     }
 
+    async fn client_key_enabled(
+        &self,
+        _: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool> {
+        Ok(false)
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        _: std::net::IpAddr,
+        _: u32,
+        _: u32,
+        _: std::time::Duration,
+    ) -> AdminStoreResult<Option<std::time::Duration>> {
+        Ok(None)
+    }
+
     async fn append_audit_event(&self, _: AdminAuditEvent) -> AdminStoreResult<()> {
         Ok(())
     }
@@ -287,6 +329,14 @@ impl AuthStore for BootstrapAuthStore {
 
 struct UnavailableStore;
 
+struct UnavailableClientKeyVerifier;
+
+impl ClientKeyVerifier for UnavailableClientKeyVerifier {
+    fn verify_client_key(&self, _: &str) -> Result<ClientApiKeyId, ClientAuthenticationError> {
+        Err(ClientAuthenticationError::InvalidKey)
+    }
+}
+
 struct UnavailableAccountGroupStore;
 
 #[async_trait]
@@ -480,6 +530,13 @@ impl AccountRuntimeStore for UnavailableStore {
 
 #[async_trait]
 impl ClientKeyStore for UnavailableStore {
+    async fn get_client_key(
+        &self,
+        _: &ClientApiKeyId,
+    ) -> AdminStoreResult<Option<ClientKeyRecord>> {
+        Err(unavailable("client key"))
+    }
+
     async fn list_client_keys(&self, _: ClientKeyListQuery) -> AdminStoreResult<ClientKeyPage> {
         Err(unavailable("client key list"))
     }

backend/crates/gateway-admin/tests/use_case/observability.rs

@@ -759,6 +759,9 @@ impl SettingsStore for FixtureSettingsStore {
             refresh_concurrency: 2,
             max_concurrent_per_account: 1,
             request_interval_ms: 0,
+            max_waiting_per_key: 0,
+            max_waiting_per_account: 0,
+            concurrency_wait_timeout_seconds: 30,
             rotation_strategy: RotationStrategy::Smart,
             min_codex_desktop_version: None,
             min_codex_cli_version: None,

backend/crates/gateway-admin/tests/use_case/settings.rs

@@ -66,6 +66,9 @@ async fn settings_should_reject_zero_refresh_margin_before_store_call() {
                 refresh_concurrency: 1,
                 max_concurrent_per_account: 1,
                 request_interval_ms: 0,
+                max_waiting_per_key: 0,
+                max_waiting_per_account: 0,
+                concurrency_wait_timeout_seconds: 30,
                 rotation_strategy: RotationStrategy::Smart,
                 min_codex_desktop_version: None,
                 min_codex_cli_version: None,

backend/crates/gateway-api/src/admin/account_groups.rs

@@ -1,5 +1,7 @@
 //! Account group HTTP wire and fixed routes.
 
+use crate::auth::SessionState;
+
 use std::collections::BTreeMap;
 
 use axum::{
@@ -22,8 +24,8 @@ use gateway_core::routing::AccountGroupId;
 use serde::{Deserialize, Serialize};
 
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
-    PageMeta, WireValidationError, wire::map_admin_service_error,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, PageMeta,
+    WireValidationError, wire::map_admin_service_error,
 };
 
 const DEFAULT_PAGE_SIZE: u32 = 50;
@@ -221,7 +223,7 @@ impl From<AccountGroupMutation> for AccountGroupMutationData {
 /// Construct all fixed account-group management routes.
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/account-groups", get(list::<S>))
@@ -238,7 +240,7 @@ async fn list<S>(
     AdminQuery(query): AdminQuery<ListAccountGroupsQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -258,7 +260,7 @@ async fn create<S>(
     AdminJson(request): AdminJson<CreateAccountGroupRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     validate_group_fields(&request.name, request.description.as_deref())?;
     mutation_response(
@@ -284,7 +286,7 @@ async fn update<S>(
     AdminJson(request): AdminJson<UpdateAccountGroupRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     validate_group_fields(&request.name, request.description.as_deref())?;
     mutation_response(
@@ -311,7 +313,7 @@ async fn enable<S>(
     AdminJson(request): AdminJson<AccountGroupIdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     set_enabled(auth, state, request, true).await
 }
@@ -322,7 +324,7 @@ async fn disable<S>(
     AdminJson(request): AdminJson<AccountGroupIdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     set_enabled(auth, state, request, false).await
 }
@@ -334,7 +336,7 @@ async fn set_enabled<S>(
     enabled: bool,
 ) -> Result<AdminResponse<AdminEnvelope<AccountGroupMutationData>>, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     mutation_response(
         StatusCode::OK,
@@ -358,7 +360,7 @@ async fn delete<S>(
     AdminJson(request): AdminJson<AccountGroupIdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     mutation_response(
         StatusCode::OK,

backend/crates/gateway-api/src/admin/accounts/credentials.rs

@@ -2,6 +2,18 @@
 
 use super::*;
 
+fn validate_account_notes(notes: Option<&str>) -> Result<(), WireValidationError> {
+    if notes.is_some_and(|notes| {
+        notes.chars().count() > 500
+            || notes
+                .chars()
+                .any(|ch| ch.is_control() && !matches!(ch, '\n' | '\r' | '\t'))
+    }) {
+        return Err(WireValidationError::new("notes"));
+    }
+    Ok(())
+}
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub(super) enum AccountProvider {
     OpenAi,
@@ -18,19 +30,22 @@ impl AccountProvider {
     }
 }
 
-/// 导入统一设置，复用编辑账号的调度和分组约束。
+/// 导入统一设置，复用编辑账号的备注、调度和分组约束。
 #[derive(Debug, Clone, Deserialize)]
 #[serde(rename_all = "camelCase", deny_unknown_fields)]
 pub struct AccountImportSettingsRequest {
+    pub notes: Option<String>,
     pub enabled: bool,
     #[serde(deserialize_with = "deserialize_required_nullable")]
     pub concurrency_limit: Option<u64>,
     pub weight: u64,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
     pub group_ids: Vec<String>,
 }
 
 impl AccountImportSettingsRequest {
     fn validate(&self) -> Result<(), WireValidationError> {
+        validate_account_notes(self.notes.as_deref())?;
         parse_concurrency_limit(self.concurrency_limit)?;
         parse_account_weight(self.weight)?;
         validate_wire_group_ids(&self.group_ids)?;
@@ -41,9 +56,11 @@ impl AccountImportSettingsRequest {
         self,
     ) -> Result<gateway_admin::model::accounts::AccountImportSettings, WireValidationError> {
         Ok(gateway_admin::model::accounts::AccountImportSettings {
+            notes: self.notes,
             enabled: self.enabled,
             concurrency_limit: parse_concurrency_limit(self.concurrency_limit)?,
             weight: parse_account_weight(self.weight)?,
+            model_access: self.model_access,
             group_ids: validate_wire_group_ids(&self.group_ids)?,
         })
     }
@@ -200,16 +217,19 @@ pub struct UpdateAccountRequest {
     pub outbound_proxy_id: Option<String>,
     pub outbound_proxy_url: Option<super::wire::AccountProxyUpdate>,
     pub account_id: String,
+    pub notes: Option<String>,
     pub enabled: bool,
     #[serde(deserialize_with = "deserialize_required_nullable")]
     pub concurrency_limit: Option<u64>,
     pub weight: u64,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
     pub group_ids: Vec<String>,
 }
 
 impl UpdateAccountRequest {
     pub fn validate(&self) -> Result<(), WireValidationError> {
         require_account_id(&self.account_id, "accountId")?;
+        validate_account_notes(self.notes.as_deref())?;
         parse_concurrency_limit(self.concurrency_limit)?;
         parse_account_weight(self.weight)?;
         validate_wire_group_ids(&self.group_ids)?;
@@ -224,9 +244,11 @@ impl UpdateAccountRequest {
                 self.outbound_proxy_url,
             )?,
             account_id: self.account_id,
+            notes: self.notes,
             enabled: self.enabled,
             concurrency_limit: parse_concurrency_limit(self.concurrency_limit)?,
             weight: parse_account_weight(self.weight)?,
+            model_access: self.model_access,
             group_ids: validate_wire_group_ids(&self.group_ids)?,
         })
     }

backend/crates/gateway-api/src/admin/accounts/handlers.rs

@@ -1,11 +1,13 @@
 //! 账号管理路由与 HTTP handler 编排。
 
+use crate::auth::SessionState;
+
 use super::*;
 
 /// 构造统一账号管理路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/accounts", get(list_accounts::<S>))
@@ -67,7 +69,7 @@ async fn batch_update_accounts<S>(
     AdminJson(request): AdminJson<BatchUpdateAccountsRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = request.into_command().map_err(map_wire_error)?;
     let result = state
@@ -88,7 +90,7 @@ async fn list_accounts<S>(
     AdminQuery(query): AdminQuery<ListQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = query.validate().map_err(map_wire_error)?;
     let page = command.page;
@@ -109,7 +111,7 @@ async fn account_detail<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -130,7 +132,7 @@ async fn export_accounts<S>(
     AdminQuery(query): AdminQuery<AccountExportQuery>,
 ) -> Result<Response, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let ids = query.into_ids().map_err(map_wire_error)?;
     let result = state
@@ -153,7 +155,7 @@ async fn import_accounts<S>(
     AdminJson(request): AdminJson<AccountImportRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let (provider, command) = request
         .into_command(auth.context().mutation_context())
@@ -181,7 +183,7 @@ async fn start_account_authorization<S>(
     AdminJson(request): AdminJson<StartAccountAuthorizationRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let (provider, command) = request
         .into_command(auth.context().mutation_context())
@@ -215,7 +217,7 @@ async fn complete_account_authorization<S>(
     AdminJson(request): AdminJson<CompleteAccountAuthorizationRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let (provider, command) = request
         .into_command(auth.context().mutation_context())
@@ -249,7 +251,7 @@ async fn rotate_account<S>(
     AdminJson(request): AdminJson<RotateAccountRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = request
         .into_command(auth.context().mutation_context())
@@ -272,7 +274,7 @@ async fn update_account<S>(
     AdminJson(request): AdminJson<UpdateAccountRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = request.into_command().map_err(map_wire_error)?;
     let result = state
@@ -293,7 +295,7 @@ async fn delete_accounts<S>(
     AdminJson(request): AdminJson<AccountDeletionRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let (provider, command) = request
         .into_command(auth.context().mutation_context())
@@ -315,7 +317,7 @@ async fn refresh_account<S>(
     AdminJson(request): AdminJson<AccountRefreshRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = request.into_command().map_err(map_wire_error)?;
     let result = state
@@ -334,7 +336,7 @@ async fn recover_account<S>(
     AdminJson(request): AdminJson<AccountActionRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = request.into_id().map_err(map_wire_error)?;
     let result = state
@@ -353,7 +355,7 @@ async fn account_quota<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -374,7 +376,7 @@ async fn account_quota_forecast<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -395,7 +397,7 @@ async fn account_personal_info<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -416,7 +418,7 @@ async fn account_profile_avatar<S>(
     AdminQuery(query): AdminQuery<AccountProfileAvatarQuery>,
 ) -> Result<Response, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let avatar = state
@@ -479,7 +481,7 @@ async fn refresh_account_quota<S>(
     AdminJson(request): AdminJson<AccountActionRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = request.into_id().map_err(map_wire_error)?;
     let result = state
@@ -500,7 +502,7 @@ async fn account_reset_credits<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -521,7 +523,7 @@ async fn consume_account_reset_credit<S>(
     AdminJson(request): AdminJson<AccountResetCreditConsumeRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = request.into_command().map_err(map_wire_error)?;
     let result = state
@@ -542,7 +544,7 @@ async fn account_models<S>(
     AdminQuery(query): AdminQuery<AccountIdQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = query.into_id().map_err(map_wire_error)?;
     let result = state
@@ -561,7 +563,7 @@ async fn refresh_account_models<S>(
     AdminJson(request): AdminJson<AccountActionRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let account_id = request.into_id().map_err(map_wire_error)?;
     let result = state
@@ -580,7 +582,7 @@ async fn test_account_connection<S>(
     AdminQuery(query): AdminQuery<AccountTestQuery>,
 ) -> Result<Sse<impl Stream<Item = Result<Event, Infallible>>>, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let (account_id, upstream_model) = query.into_command().map_err(map_wire_error)?;
     let stream = state

backend/crates/gateway-api/src/admin/accounts/mod.rs

@@ -46,8 +46,8 @@ use uuid::{Uuid, Version};
 
 use super::presenter::{format_compact_number, format_decimal_currency, format_number};
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
-    PageMeta, WireValidationError,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, PageMeta,
+    WireValidationError,
 };
 
 const DEFAULT_PAGE_SIZE: u32 = 50;
@@ -75,8 +75,8 @@ pub(super) use presenter::quota_window_view;
 pub use wire::*;
 
 use credentials::{
-    AccountProvider, deserialize_required_nullable, parse_account_weight, parse_concurrency_limit,
-    provider_document_value, require_account_id, validate_wire_group_ids,
+    AccountProvider, parse_account_weight, parse_concurrency_limit, provider_document_value,
+    require_account_id, validate_wire_group_ids,
 };
 use presenter::*;
 use wire::BatchUpdatedAccountsData;

backend/crates/gateway-api/src/admin/accounts/presenter.rs

@@ -75,6 +75,7 @@ pub(super) fn account_view(item: AccountDirectoryItem, now: DateTime<Utc>) -> Ac
     AccountView {
         id: account.id.clone(),
         name: account.name,
+        notes: account.notes,
         provider: account.provider_kind.to_string(),
         groups: account
             .groups
@@ -103,6 +104,7 @@ pub(super) fn account_view(item: AccountDirectoryItem, now: DateTime<Utc>) -> Ac
         enabled: account.enabled,
         concurrency_limit: account.concurrency_limit.map(|limit| limit.get()),
         weight: account.weight.get(),
+        model_access: account.model_access,
         outbound_proxy_endpoint: account
             .outbound_proxy
             .as_ref()

backend/crates/gateway-api/src/admin/accounts/wire.rs

@@ -64,11 +64,12 @@ pub struct BatchUpdateAccountsRequest {
     pub outbound_proxy_id: Option<String>,
     pub outbound_proxy_url: Option<AccountProxyUpdate>,
     pub account_ids: Vec<String>,
-    pub enabled: bool,
-    #[serde(deserialize_with = "deserialize_required_nullable")]
-    pub concurrency_limit: Option<u64>,
-    pub weight: u64,
-    pub group_ids: Vec<String>,
+    pub enabled: Option<bool>,
+    #[serde(default, deserialize_with = "deserialize_optional_nullable_limit")]
+    pub concurrency_limit: Option<Option<u64>>,
+    pub weight: Option<u64>,
+    pub model_access: Option<gateway_core::account::AccountModelAccess>,
+    pub group_ids: Option<Vec<String>>,
 }
 
 #[derive(Debug, Clone, Serialize)]
@@ -91,6 +92,12 @@ impl From<AccountsUpdateResult> for BatchUpdatedAccountsData {
     }
 }
 
+fn deserialize_optional_nullable_limit<'de, D: serde::Deserializer<'de>>(
+    deserializer: D,
+) -> Result<Option<Option<u64>>, D::Error> {
+    Option::<u64>::deserialize(deserializer).map(Some)
+}
+
 impl BatchUpdateAccountsRequest {
     pub fn validate(&self) -> Result<(), WireValidationError> {
         if self.account_ids.is_empty()
@@ -103,9 +110,23 @@ impl BatchUpdateAccountsRequest {
         {
             return Err(WireValidationError::new("accountIds"));
         }
-        validate_wire_group_ids(&self.group_ids)?;
-        parse_concurrency_limit(self.concurrency_limit)?;
-        parse_account_weight(self.weight)?;
+        if let Some(group_ids) = &self.group_ids {
+            validate_wire_group_ids(group_ids)?;
+        }
+        self.concurrency_limit
+            .map(parse_concurrency_limit)
+            .transpose()?;
+        self.weight.map(parse_account_weight).transpose()?;
+        if self.enabled.is_none()
+            && self.concurrency_limit.is_none()
+            && self.weight.is_none()
+            && self.group_ids.is_none()
+            && self.model_access.is_none()
+            && self.outbound_proxy_id.is_none()
+            && self.outbound_proxy_url.is_none()
+        {
+            return Err(WireValidationError::new("accountSettings"));
+        }
         Ok(())
     }
 
@@ -115,9 +136,17 @@ impl BatchUpdateAccountsRequest {
             outbound_proxy: proxy_selection(self.outbound_proxy_id, self.outbound_proxy_url)?,
             account_ids: self.account_ids,
             enabled: self.enabled,
-            concurrency_limit: parse_concurrency_limit(self.concurrency_limit)?,
-            weight: parse_account_weight(self.weight)?,
-            group_ids: validate_wire_group_ids(&self.group_ids)?,
+            concurrency_limit: self
+                .concurrency_limit
+                .map(parse_concurrency_limit)
+                .transpose()?,
+            weight: self.weight.map(parse_account_weight).transpose()?,
+            model_access: self.model_access,
+            group_ids: self
+                .group_ids
+                .as_deref()
+                .map(validate_wire_group_ids)
+                .transpose()?,
         })
     }
 }
@@ -240,6 +269,7 @@ pub struct AccountView {
     pub outbound_proxy_endpoint: Option<String>,
     pub id: String,
     pub name: String,
+    pub notes: Option<String>,
     pub provider: String,
     pub groups: Vec<AccountGroupRefView>,
     pub resource_ref: String,
@@ -260,6 +290,7 @@ pub struct AccountView {
     pub enabled: bool,
     pub concurrency_limit: Option<u32>,
     pub weight: u16,
+    pub model_access: gateway_core::account::AccountModelAccess,
     pub access_token_expires_at: Option<String>,
     pub access_token_expires_at_display: Option<String>,
     pub refresh_token_expires_at: Option<String>,

backend/crates/gateway-api/src/admin/auth.rs

@@ -1,54 +1,17 @@
-//! 管理员登录、会话与管理请求鉴权接线。
-
-use std::fmt;
+//! 管理端权限校验与请求审计上下文。
 
 use axum::{
-    Router,
-    extract::{FromRequestParts, State},
-    http::{
-        HeaderMap, HeaderValue, StatusCode,
-        header::{ORIGIN, SET_COOKIE},
-        request::Parts,
-    },
-    response::{IntoResponse, Response},
-    routing::{get, post},
+    extract::FromRequestParts,
+    http::{HeaderMap, request::Parts},
 };
-use gateway_admin::{
-    AdminServices,
-    model::auth::{AdminPrincipal, AdminRequestContext, LoginCommand, LoginError},
-};
-use serde::{Deserialize, Serialize};
+use gateway_admin::model::auth::{AdminPrincipal, AdminRequestContext};
 use tower_http::request_id::RequestId;
-use url::Url;
 
-use super::{AdminEnvelope, AdminError, AdminJson, AdminResponse};
+use crate::{auth::SessionState, session_cookie};
+
+use super::{AdminError, wire::map_admin_service_error};
 
 const REQUEST_ID_HEADER: &str = "x-request-id";
-const ADMIN_SESSION_COOKIE: &str = "cpr_admin_session";
-
-/// 所有管理 HTTP 模块从 state 消费同一个认证用例端口。
-pub trait AdminSessionState {
-    fn admin_services(&self) -> &AdminServices;
-}
-
-fn admin_session_cookie_attrs(headers: &HeaderMap) -> &'static str {
-    // 同源管理端的 POST 由浏览器携带 Origin，HTTPS 反代回源 HTTP 不改变它。
-    // 只对明确的 HTTP 来源放宽；缺失、opaque 或非法来源继续使用 Secure。
-    let http_origin = headers.get_all(ORIGIN).iter().count() == 1
-        && headers
-            .get(ORIGIN)
-            .and_then(|value| value.to_str().ok())
-            .is_some_and(|origin| {
-                Url::parse(origin).is_ok_and(|url| {
-                    url.scheme() == "http" && url.origin().ascii_serialization() == origin
-                })
-            });
-    if http_origin {
-        "Path=/; HttpOnly; SameSite=Lax"
-    } else {
-        "Path=/; Secure; HttpOnly; SameSite=Lax"
-    }
-}
 
 /// 已通过管理员会话或部署级管理 API Key 鉴权的请求。
 pub struct AdminAuth {
@@ -64,7 +27,7 @@ impl AdminAuth {
 
 impl<S> FromRequestParts<S> for AdminAuth
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     type Rejection = AdminError;
 
@@ -97,23 +60,23 @@ fn admin_request_id(parts: &Parts) -> Option<String> {
 
 pub async fn require_admin_session<S>(state: &S, headers: &HeaderMap) -> Result<String, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     match state
         .admin_services()
         .auth()
-        .resolve_admin_user_id(admin_session_cookie(headers).as_deref())
+        .resolve_admin_user_id(session_cookie::value(headers).as_deref())
         .await
     {
         Ok(Some(admin_user_id)) => Ok(admin_user_id),
-        Ok(None) => Err(AdminError::admin_session_required()),
-        Err(_) => Err(AdminError::internal()),
+        Ok(None) => Err(AdminError::session_required()),
+        Err(error) => Err(map_admin_service_error(error)),
     }
 }
 
 async fn require_admin_auth<S>(state: &S, headers: &HeaderMap) -> Result<AdminPrincipal, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     if let Some(api_key) = admin_api_key_header(headers) {
         return match state
@@ -124,7 +87,7 @@ where
         {
             Ok(true) => Ok(AdminPrincipal::ApiKey),
             Ok(false) => Err(AdminError::invalid_admin_api_key()),
-            Err(_) => Err(AdminError::internal()),
+            Err(error) => Err(map_admin_service_error(error)),
         };
     }
 
@@ -137,176 +100,3 @@ fn admin_api_key_header(headers: &HeaderMap) -> Option<String> {
     let value = headers.get("x-api-key")?.to_str().ok()?.trim();
     (!value.is_empty()).then(|| value.to_owned())
 }
-
-/// 管理员登录请求；密码不得进入 Debug。
-#[derive(Deserialize)]
-#[serde(rename_all = "camelCase", deny_unknown_fields)]
-pub struct AdminLoginRequest {
-    username: Option<String>,
-    password: String,
-}
-
-impl AdminLoginRequest {
-    /// 取出认证用例所需字段。
-    #[must_use]
-    pub fn into_parts(self) -> (Option<String>, String) {
-        (self.username, self.password)
-    }
-}
-
-impl fmt::Debug for AdminLoginRequest {
-    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
-        formatter
-            .debug_struct("AdminLoginRequest")
-            .field("username", &self.username)
-            .field("password", &"[REDACTED]")
-            .finish()
-    }
-}
-
-/// 登录成功响应。
-#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
-#[serde(rename_all = "camelCase")]
-pub struct AdminLoginData {
-    expires_at: String,
-}
-
-impl AdminLoginData {
-    #[must_use]
-    pub fn new(expires_at: String) -> Self {
-        Self { expires_at }
-    }
-}
-
-/// 管理员会话状态响应。
-#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
-#[serde(rename_all = "camelCase")]
-pub struct AdminSessionStatusData {
-    authenticated: bool,
-}
-
-impl AdminSessionStatusData {
-    #[must_use]
-    pub const fn new(authenticated: bool) -> Self {
-        Self { authenticated }
-    }
-}
-
-/// 登出成功响应。
-#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
-pub struct AdminLogoutData {
-    message: &'static str,
-}
-
-impl AdminLogoutData {
-    #[must_use]
-    pub const fn new() -> Self {
-        Self {
-            message: "Logged out successfully",
-        }
-    }
-}
-
-impl Default for AdminLogoutData {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
-/// 构造固定 GET/POST 管理员认证路由。
-pub fn router<S>() -> Router<S>
-where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
-{
-    Router::new()
-        .route("/api/admin/auth/login", post(login::<S>))
-        .route("/api/admin/auth/status", get(session_status::<S>))
-        .route("/api/admin/auth/logout", post(logout::<S>))
-}
-
-async fn login<S>(
-    State(state): State<S>,
-    headers: HeaderMap,
-    AdminJson(payload): AdminJson<AdminLoginRequest>,
-) -> Result<Response, AdminError>
-where
-    S: AdminSessionState + Send + Sync,
-{
-    let (username, password) = payload.into_parts();
-    let session = state
-        .admin_services()
-        .auth()
-        .login(LoginCommand { username, password })
-        .await
-        .map_err(map_login_error)?;
-    let mut response = AdminResponse::new(
-        StatusCode::OK,
-        AdminEnvelope::ok(AdminLoginData::new(session.expires_at.to_rfc3339())),
-    )
-    .into_response();
-    let cookie = format!(
-        "{ADMIN_SESSION_COOKIE}={}; {}",
-        session.session_id,
-        admin_session_cookie_attrs(&headers)
-    );
-    response.headers_mut().insert(
-        SET_COOKIE,
-        HeaderValue::from_str(&cookie).map_err(|_| AdminError::internal())?,
-    );
-    Ok(response)
-}
-
-async fn session_status<S>(
-    State(state): State<S>,
-    headers: HeaderMap,
-) -> Result<impl IntoResponse, AdminError>
-where
-    S: AdminSessionState + Send + Sync,
-{
-    let authenticated = state
-        .admin_services()
-        .auth()
-        .validate_session(admin_session_cookie(&headers).as_deref())
-        .await
-        .map_err(|_| AdminError::internal())?;
-    Ok(AdminResponse::new(
-        StatusCode::OK,
-        AdminEnvelope::ok(AdminSessionStatusData::new(authenticated)),
-    ))
-}
-
-async fn logout<S>(State(state): State<S>, headers: HeaderMap) -> Result<Response, AdminError>
-where
-    S: AdminSessionState + Send + Sync,
-{
-    if let Some(session_id) = admin_session_cookie(&headers) {
-        let _ = state.admin_services().auth().logout(&session_id).await;
-    }
-    let mut response =
-        AdminResponse::new(StatusCode::OK, AdminEnvelope::ok(AdminLogoutData::new()))
-            .into_response();
-    let cookie = format!(
-        "{ADMIN_SESSION_COOKIE}=; {}; Max-Age=0",
-        admin_session_cookie_attrs(&headers)
-    );
-    response.headers_mut().insert(
-        SET_COOKIE,
-        HeaderValue::from_str(&cookie).map_err(|_| AdminError::internal())?,
-    );
-    Ok(response)
-}
-
-fn admin_session_cookie(headers: &HeaderMap) -> Option<String> {
-    let cookie = headers.get("cookie")?.to_str().ok()?;
-    cookie.split(';').find_map(|part| {
-        let (name, value) = part.trim().split_once('=')?;
-        (name == ADMIN_SESSION_COOKIE).then(|| value.to_owned())
-    })
-}
-
-fn map_login_error(error: LoginError) -> AdminError {
-    match error {
-        LoginError::InvalidCredentials => AdminError::invalid_admin_credentials(),
-        LoginError::Unavailable => AdminError::internal(),
-    }
-}

backend/crates/gateway-api/src/admin/backups.rs

@@ -1,7 +1,10 @@
 //! 备份设置、计划、手动创建、记录、下载与删除路由。
+
 //!
 //! 路径统一位于 `/api/admin/settings/backups/*`，内部由独立 `BackupService` 承担业务。
 
+use crate::auth::SessionState;
+
 use axum::{
     Router,
     extract::State,
@@ -19,7 +22,7 @@ use serde::{Deserialize, Serialize};
 
 use super::{
     AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminPageData, AdminQuery, AdminResponse,
-    AdminSessionState, PageMeta, wire::map_admin_service_error,
+    PageMeta, wire::map_admin_service_error,
 };
 
 /// 备份设置视图；`secretAccessKey` 返回已保存的明文凭据，由前端掩码显示。
@@ -200,7 +203,7 @@ impl From<ConnectionTestResult> for ConnectionTestView {
 /// 构造固定 GET/POST 备份路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/settings/backups", get(backup_settings::<S>))
@@ -239,7 +242,7 @@ async fn backup_settings<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let settings = state
         .admin_services()
@@ -259,7 +262,7 @@ async fn update_backup_storage<S>(
     AdminJson(request): AdminJson<UpdateBackupStorageRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = UpdateBackupStorageCommand {
         endpoint: request.endpoint,
@@ -287,7 +290,7 @@ async fn test_backup_storage<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -307,7 +310,7 @@ async fn update_backup_schedule<S>(
     AdminJson(request): AdminJson<UpdateBackupScheduleRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = UpdateBackupScheduleCommand {
         schedule_enabled: request.schedule_enabled,
@@ -334,7 +337,7 @@ async fn backup_records<S>(
     AdminQuery(query): AdminQuery<BackupRecordsQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let page_size = gateway_admin::model::PageSize::new(query.page_size.unwrap_or(20))
         .map_err(|_| AdminError::bad_request("pageSize 不合法"))?;
@@ -389,7 +392,7 @@ async fn create_backup<S>(
     body: Option<AdminJson<CreateBackupRequest>>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let expires_at = body
         .and_then(|body| body.0.expires_in_days)
@@ -420,7 +423,7 @@ async fn download_backup_url<S>(
     AdminJson(request): AdminJson<BackupIdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -440,7 +443,7 @@ async fn delete_backup<S>(
     AdminJson(request): AdminJson<BackupIdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let record = state
         .admin_services()

backend/crates/gateway-api/src/admin/client_keys.rs

@@ -1,5 +1,7 @@
 //! Client API Key 管理 wire contract。
 
+use crate::auth::SessionState;
+
 use std::fmt;
 
 use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD};
@@ -27,7 +29,7 @@ use axum::{
 };
 
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse,
     WireValidationError, wire::map_admin_service_error,
 };
 
@@ -679,7 +681,7 @@ fn validate_optional_text(
 /// 构造固定 GET/POST 且 ID 仅位于 query/body 的 Client API Key 路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/client-keys", get(list_client_keys::<S>))
@@ -712,7 +714,7 @@ async fn list_client_keys<S>(
     AdminQuery(query): AdminQuery<ListClientKeysQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -730,7 +732,7 @@ async fn create_client_key<S>(
     AdminJson(payload): AdminJson<CreateClientKeyRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = payload.into_command().map_err(map_wire_error)?;
     let result = state
@@ -751,7 +753,7 @@ async fn reveal_client_key<S>(
     AdminQuery(query): AdminQuery<ClientKeyIdQuery>,
 ) -> Result<Response, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let id = query.into_domain_id().map_err(map_wire_error)?;
     let result = state
@@ -777,7 +779,7 @@ async fn update_client_key<S>(
     AdminJson(payload): AdminJson<UpdateClientKeyRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = payload.into_command().map_err(map_wire_error)?;
     mutation_response(
@@ -795,7 +797,7 @@ async fn disable_client_key<S>(
     AdminJson(payload): AdminJson<ClientKeyMutationRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let id = payload.into_domain_id().map_err(map_wire_error)?;
     mutation_response(
@@ -816,7 +818,7 @@ async fn enable_client_key<S>(
     AdminJson(payload): AdminJson<ClientKeyMutationRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let id = payload.into_domain_id().map_err(map_wire_error)?;
     mutation_response(
@@ -837,7 +839,7 @@ async fn delete_client_key<S>(
     AdminJson(payload): AdminJson<ClientKeyMutationRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let id = payload.into_domain_id().map_err(map_wire_error)?;
     mutation_response(

backend/crates/gateway-api/src/admin/mod.rs

@@ -1,5 +1,7 @@
 //! 管理端 HTTP adapter、wire contract 与固定路由。
 
+use crate::auth::SessionState;
+
 use axum::{
     Router,
     http::{HeaderValue, header},
@@ -21,7 +23,7 @@ pub mod settings;
 pub mod system;
 pub mod wire;
 
-pub use auth::{AdminAuth, AdminSessionState};
+pub use auth::AdminAuth;
 pub use extract::{AdminJson, AdminQuery};
 pub use wire::{
     ADMIN_OK_CODE, ADMIN_OK_MESSAGE, AdminEnvelope, AdminError, AdminErrorBody, AdminErrorCode,
@@ -31,13 +33,12 @@ pub use wire::{
 /// 构造完整且固定的 `/api/admin` 路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .merge(account_groups::router::<S>())
         .merge(proxies::router::<S>())
         .merge(accounts::router::<S>())
-        .merge(auth::router::<S>())
         .merge(backups::router::<S>())
         .merge(client_keys::router::<S>())
         .merge(observability::router::<S>())

backend/crates/gateway-api/src/admin/observability/mod.rs

@@ -10,8 +10,8 @@ use serde_json::Value;
 
 use super::presenter::{format_compact_number, format_decimal_currency, format_number};
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminQuery, AdminResponse, AdminSessionState,
-    WireValidationError, wire::map_admin_service_error,
+    AdminAuth, AdminEnvelope, AdminError, AdminQuery, AdminResponse, WireValidationError,
+    wire::map_admin_service_error,
 };
 
 mod presenter;

backend/crates/gateway-api/src/admin/observability/routes.rs

@@ -1,10 +1,12 @@
 //! 固定 `/api/admin` 观测路由与 handler。
 
+use crate::auth::SessionState;
+
 use super::*;
 
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/dashboard/summary", get(dashboard_summary::<S>))
@@ -35,7 +37,7 @@ pub(crate) async fn dashboard_summary<S>(
     AdminQuery(query): AdminQuery<DashboardQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let kind = query.trend_kind().map_err(map_wire_error)?;
     // 概览默认按中国时区当日统计，与单独趋势接口保持同一口径。
@@ -59,7 +61,7 @@ pub(crate) async fn dashboard_trend<S>(
     AdminQuery(query): AdminQuery<DashboardQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let kind = query.trend_kind().map_err(map_wire_error)?;
     let range = dashboard_today_range(query.start_time.as_deref(), query.end_time.as_deref())
@@ -82,7 +84,7 @@ pub(crate) async fn usage_records<S>(
     AdminQuery(query): AdminQuery<UsageQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = usage_command(&query).map_err(map_wire_error)?;
     let result = state
@@ -101,7 +103,7 @@ pub(crate) async fn usage_record_detail<S>(
     AdminQuery(query): AdminQuery<DetailQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     query.validate().map_err(map_wire_error)?;
     let result = state
@@ -122,7 +124,7 @@ pub(crate) async fn usage_records_summary<S>(
     AdminQuery(query): AdminQuery<UsageQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let range = usage_range(query.start_time.as_deref(), query.end_time.as_deref())
         .map_err(map_wire_error)?;
@@ -145,7 +147,7 @@ pub(crate) async fn usage_insights_overview<S>(
     AdminQuery(query): AdminQuery<UsageQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let range = usage_range(query.start_time.as_deref(), query.end_time.as_deref())
         .map_err(map_wire_error)?;
@@ -168,7 +170,7 @@ pub(crate) async fn usage_insights_diagnostics<S>(
     AdminQuery(query): AdminQuery<DiagnosticsQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let dimension = query.dimension().map_err(map_wire_error)?;
     let range = usage_range(query.start_time.as_deref(), query.end_time.as_deref())
@@ -198,7 +200,7 @@ pub(crate) async fn ops_errors<S>(
     AdminQuery(query): AdminQuery<OpsQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = ops_command(&query).map_err(map_wire_error)?;
     let result = state

backend/crates/gateway-api/src/admin/proxies.rs

@@ -1,3 +1,5 @@
+use crate::auth::SessionState;
+
 use axum::{
     Router,
     extract::State,
@@ -15,8 +17,7 @@ use gateway_admin::model::{
 use serde::{Deserialize, Serialize};
 
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
-    PageMeta,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, PageMeta,
     accounts::{AccountGroupRefView, AccountProxyUpdate},
 };
 
@@ -170,7 +171,7 @@ impl From<ProxyMutation> for MutationView {
 
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/proxies", get(list::<S>))
@@ -203,7 +204,7 @@ async fn list<S>(
     AdminQuery(query): AdminQuery<ListQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -237,7 +238,7 @@ async fn list_accounts<S>(
     AdminQuery(query): AdminQuery<AccountsQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -295,7 +296,7 @@ async fn create<S>(
     AdminJson(request): AdminJson<CreateRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let proxy = request
         .proxy_url
@@ -325,7 +326,7 @@ async fn remove_account<S>(
     AdminJson(request): AdminJson<RemoveAccountRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let revision = state
         .admin_services()
@@ -349,7 +350,7 @@ async fn update<S>(
     AdminJson(request): AdminJson<UpdateRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let proxy = request
         .proxy_url
@@ -385,7 +386,7 @@ async fn delete<S>(
     AdminJson(request): AdminJson<IdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -409,7 +410,7 @@ async fn probe<S>(
     AdminJson(request): AdminJson<ProbeRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let proxy = request
         .proxy_url
@@ -433,7 +434,7 @@ async fn test<S>(
     AdminJson(request): AdminJson<IdRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()

backend/crates/gateway-api/src/admin/settings.rs

@@ -1,5 +1,7 @@
 //! Runtime settings、旧设置页聚合投影与明文 Admin API Key wire。
 
+use crate::auth::SessionState;
+
 use std::{collections::BTreeMap, fmt};
 
 use axum::{
@@ -21,7 +23,7 @@ use gateway_core::routing::{PublicModelId, UpstreamModelId};
 use serde::{Deserialize, Serialize};
 
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse,
     WireValidationError, wire::map_admin_service_error,
 };
 
@@ -37,6 +39,9 @@ pub struct RuntimeSettingsView {
     pub refresh_concurrency: u64,
     pub max_concurrent_per_account: u64,
     pub request_interval_ms: u64,
+    pub max_waiting_per_key: u32,
+    pub max_waiting_per_account: u32,
+    pub concurrency_wait_timeout_seconds: u32,
     pub rotation_strategy: String,
     pub min_codex_desktop_version: Option<String>,
     pub min_codex_cli_version: Option<String>,
@@ -55,6 +60,9 @@ pub struct UpdateRuntimeSettingsRequest {
     pub refresh_concurrency: u64,
     pub max_concurrent_per_account: u64,
     pub request_interval_ms: u64,
+    pub max_waiting_per_key: u32,
+    pub max_waiting_per_account: u32,
+    pub concurrency_wait_timeout_seconds: u32,
     pub rotation_strategy: String,
     pub min_codex_desktop_version: Option<String>,
     pub min_codex_cli_version: Option<String>,
@@ -67,6 +75,17 @@ impl UpdateRuntimeSettingsRequest {
     /// 校验公共运行参数。
     pub fn validate(&self) -> Result<(), WireValidationError> {
         validate_model_mappings(&self.model_mappings)?;
+        for (value, field) in [
+            (self.max_waiting_per_key, "maxWaitingPerKey"),
+            (self.max_waiting_per_account, "maxWaitingPerAccount"),
+        ] {
+            if value > 1_000 {
+                return Err(WireValidationError::new(field));
+            }
+        }
+        if !(1..=120).contains(&self.concurrency_wait_timeout_seconds) {
+            return Err(WireValidationError::new("concurrencyWaitTimeoutSeconds"));
+        }
         for (value, field) in [
             (self.refresh_margin_seconds, "refreshMarginSeconds"),
             (self.refresh_concurrency, "refreshConcurrency"),
@@ -104,6 +123,9 @@ impl UpdateRuntimeSettingsRequest {
             max_concurrent_per_account: u32::try_from(self.max_concurrent_per_account)
                 .map_err(|_| WireValidationError::new("settingsMaxConcurrencyOverflow"))?,
             request_interval_ms: self.request_interval_ms,
+            max_waiting_per_key: self.max_waiting_per_key,
+            max_waiting_per_account: self.max_waiting_per_account,
+            concurrency_wait_timeout_seconds: self.concurrency_wait_timeout_seconds,
             rotation_strategy: RotationStrategy::parse(&self.rotation_strategy)
                 .ok_or_else(|| WireValidationError::new("rotationStrategy"))?,
             min_codex_desktop_version: self.min_codex_desktop_version,
@@ -126,6 +148,9 @@ impl From<RuntimeSettings> for RuntimeSettingsView {
             refresh_concurrency: u64::from(settings.refresh_concurrency),
             max_concurrent_per_account: u64::from(settings.max_concurrent_per_account),
             request_interval_ms: settings.request_interval_ms,
+            max_waiting_per_key: settings.max_waiting_per_key,
+            max_waiting_per_account: settings.max_waiting_per_account,
+            concurrency_wait_timeout_seconds: settings.concurrency_wait_timeout_seconds,
             rotation_strategy: settings.rotation_strategy.as_str().to_owned(),
             min_codex_desktop_version: settings.min_codex_desktop_version,
             min_codex_cli_version: settings.min_codex_cli_version,
@@ -228,7 +253,7 @@ impl Default for DeletedAdminApiKey {
 /// 构造固定 GET/POST 设置路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/settings", get(settings::<S>))
@@ -257,7 +282,7 @@ async fn codex_desktop_windows_downloads<S>(
     AdminQuery(query): AdminQuery<ClientDownloadsQuery>,
 ) -> impl IntoResponse
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let downloads = state
         .admin_services()
@@ -275,7 +300,7 @@ async fn settings<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -295,7 +320,7 @@ async fn update_settings<S>(
     AdminJson(request): AdminJson<UpdateRuntimeSettingsRequest>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let command = request.into_command().map_err(map_wire_error)?;
     let result = state
@@ -315,7 +340,7 @@ async fn admin_api_key_status<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let exists = state
         .admin_services()
@@ -334,7 +359,7 @@ async fn regenerate_admin_api_key<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -355,7 +380,7 @@ async fn delete_admin_api_key<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     state
         .admin_services()

backend/crates/gateway-api/src/admin/system.rs

@@ -1,7 +1,10 @@
 //! 系统管理接口的查询与请求 wire contract。
+
 //!
 //! 这里不依赖更新服务或进程控制；应用层通过窄端口提供系统操作事实。
 
+use crate::auth::SessionState;
+
 use std::convert::Infallible;
 
 use axum::{
@@ -23,7 +26,7 @@ use gateway_admin::model::system::{
 use serde::{Deserialize, Serialize};
 
 use super::{
-    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse, AdminSessionState,
+    AdminAuth, AdminEnvelope, AdminError, AdminJson, AdminQuery, AdminResponse,
     wire::map_admin_service_error,
 };
 
@@ -214,7 +217,7 @@ struct SystemUpdateEventView {
 /// 构造固定 GET/POST 系统管理路由。
 pub fn router<S>() -> Router<S>
 where
-    S: AdminSessionState + Clone + Send + Sync + 'static,
+    S: SessionState + Clone + Send + Sync + 'static,
 {
     Router::new()
         .route("/api/admin/system/version", get(version::<S>))
@@ -234,7 +237,7 @@ async fn version<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let version = state
         .admin_services()
@@ -254,7 +257,7 @@ async fn update_detail<S>(
     AdminQuery(query): AdminQuery<UpdateDetailQuery>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let detail = state
         .admin_services()
@@ -273,7 +276,7 @@ async fn update_event_stream<S>(
     State(state): State<S>,
 ) -> Result<Sse<impl Stream<Item = Result<Event, Infallible>>>, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let stream = state
         .admin_services()
@@ -293,7 +296,7 @@ async fn perform_update<S>(
     payload: Option<AdminJson<UpdateRequest>>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -328,7 +331,7 @@ async fn update_status<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let status = state
         .admin_services()
@@ -347,7 +350,7 @@ async fn rollback<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()
@@ -378,7 +381,7 @@ async fn restart<S>(
     State(state): State<S>,
 ) -> Result<impl IntoResponse, AdminError>
 where
-    S: AdminSessionState + Send + Sync,
+    S: SessionState + Send + Sync,
 {
     let result = state
         .admin_services()

backend/crates/gateway-api/src/admin/wire.rs

@@ -96,17 +96,19 @@ impl AdminErrorCode {
     pub const INVALID_TIME_RANGE: Self = Self(40002);
     /// 模型来源不合法。
     pub const INVALID_MODEL_SOURCE: Self = Self(40003);
-    /// 缺少管理员会话。
+    /// 缺少登录会话。
     pub const SESSION_REQUIRED: Self = Self(40101);
-    /// 管理员登录凭据错误。
+    /// 登录凭据错误。
     pub const INVALID_CREDENTIALS: Self = Self(40102);
     /// 管理 API Key 错误。
     pub const INVALID_API_KEY: Self = Self(40103);
+    /// 当前会话无权访问。
+    pub const FORBIDDEN: Self = Self(40301);
     /// 资源不存在。
     pub const NOT_FOUND: Self = Self(40401);
     /// 配置 revision 或资源状态冲突。
     pub const CONFLICT: Self = Self(40901);
-    /// 管理员登录尝试过多。
+    /// 登录尝试过多。
     pub const TOO_MANY_LOGIN_ATTEMPTS: Self = Self(42901);
     /// 设置持久化失败。
     pub const SETTINGS_PERSIST: Self = Self(50000);
@@ -165,6 +167,7 @@ impl AdminErrorBody {
 pub struct AdminError {
     status: StatusCode,
     body: AdminErrorBody,
+    retry_after_seconds: Option<u64>,
 }
 
 #[derive(Debug, Clone, Copy)]
@@ -202,12 +205,12 @@ const INVALID_TIME_RANGE: AdminErrorSpec = AdminErrorSpec::new(
 const SESSION_REQUIRED: AdminErrorSpec = AdminErrorSpec::new(
     StatusCode::UNAUTHORIZED,
     AdminErrorCode::SESSION_REQUIRED,
-    "需要管理员登录",
+    "需要登录",
 );
 const INVALID_CREDENTIALS: AdminErrorSpec = AdminErrorSpec::new(
     StatusCode::UNAUTHORIZED,
     AdminErrorCode::INVALID_CREDENTIALS,
-    "管理员用户名或密码错误",
+    "登录凭据错误",
 );
 const INVALID_API_KEY: AdminErrorSpec = AdminErrorSpec::new(
     StatusCode::UNAUTHORIZED,
@@ -255,6 +258,7 @@ impl AdminError {
         Self {
             status,
             body: AdminErrorBody::new(code, message),
+            retry_after_seconds: None,
         }
     }
 
@@ -286,11 +290,11 @@ impl AdminError {
         Self::new(BAD_REQUEST.status, BAD_REQUEST.code, message)
     }
 
-    pub fn admin_session_required() -> Self {
+    pub fn session_required() -> Self {
         Self::from_spec(SESSION_REQUIRED)
     }
 
-    pub fn invalid_admin_credentials() -> Self {
+    pub fn invalid_credentials() -> Self {
         Self::from_spec(INVALID_CREDENTIALS)
     }
 
@@ -302,6 +306,10 @@ impl AdminError {
         Self::from_spec(TOO_MANY_LOGIN_ATTEMPTS)
     }
 
+    pub fn forbidden() -> Self {
+        Self::new(StatusCode::FORBIDDEN, AdminErrorCode::FORBIDDEN, "无权访问")
+    }
+
     pub fn conflict(message: impl Into<String>) -> Self {
         Self::new(StatusCode::CONFLICT, AdminErrorCode::CONFLICT, message)
     }
@@ -325,6 +333,12 @@ impl AdminError {
     pub fn service_unavailable() -> Self {
         Self::from_spec(SERVICE_UNAVAILABLE)
     }
+
+    #[must_use]
+    pub fn with_retry_after(mut self, seconds: u64) -> Self {
+        self.retry_after_seconds = Some(seconds.max(1));
+        self
+    }
 }
 
 /// 把管理用例的稳定错误分类映射到既有 HTTP 错误 contract。
@@ -333,7 +347,8 @@ pub(crate) fn map_admin_service_error(error: gateway_admin::model::AdminError) -
 
     let mut response = match error.kind() {
         AdminErrorKind::Invalid => AdminError::bad_request(error.message()),
-        AdminErrorKind::Unauthorized => AdminError::admin_session_required(),
+        AdminErrorKind::Forbidden => AdminError::forbidden(),
+        AdminErrorKind::Unauthorized => AdminError::session_required(),
         AdminErrorKind::NotFound => AdminError::not_found(error.message()),
         AdminErrorKind::Conflict => AdminError::conflict(error.message()),
         AdminErrorKind::RateLimited => AdminError::too_many_login_attempts(),
@@ -358,7 +373,15 @@ pub(crate) fn map_admin_service_error(error: gateway_admin::model::AdminError) -
 
 impl IntoResponse for AdminError {
     fn into_response(self) -> Response {
-        (self.status, Json(self.body)).into_response()
+        let mut response = (self.status, Json(self.body)).into_response();
+        if let Some(seconds) = self.retry_after_seconds
+            && let Ok(value) = axum::http::HeaderValue::from_str(&seconds.to_string())
+        {
+            response
+                .headers_mut()
+                .insert(axum::http::header::RETRY_AFTER, value);
+        }
+        response
     }
 }
 

backend/crates/gateway-api/src/auth.rs

@@ -0,0 +1,232 @@
+//! 控制面统一登录、会话恢复与退出；身份由认证用例返回。
+
+use std::{fmt, net::SocketAddr};
+
+use axum::{
+    Router,
+    extract::{ConnectInfo, State},
+    http::{HeaderMap, HeaderValue, StatusCode, header},
+    middleware,
+    response::{IntoResponse, Response},
+    routing::{any, get, post},
+};
+use gateway_admin::{
+    AdminServices,
+    model::auth::{AuthSession, LoginCommand, LoginError, SessionSubject},
+};
+use serde::{Deserialize, Serialize};
+
+use crate::{
+    admin::{AdminEnvelope, AdminError, AdminJson, AdminResponse, wire::map_admin_service_error},
+    session_cookie,
+};
+
+/// 控制面 HTTP adapter 消费同一组用例；权限由各入口服务端校验。
+pub trait SessionState {
+    fn admin_services(&self) -> &AdminServices;
+}
+
+#[derive(Deserialize)]
+#[serde(tag = "mode", rename_all = "camelCase", deny_unknown_fields)]
+pub enum LoginRequest {
+    Admin {
+        username: Option<String>,
+        password: String,
+    },
+    Key {
+        #[serde(rename = "apiKey")]
+        api_key: String,
+    },
+}
+
+impl fmt::Debug for LoginRequest {
+    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            Self::Admin { username, .. } => formatter
+                .debug_struct("AdminLogin")
+                .field("username", username)
+                .field("password", &"[REDACTED]")
+                .finish(),
+            Self::Key { .. } => formatter
+                .debug_struct("KeyLogin")
+                .field("api_key", &"[REDACTED]")
+                .finish(),
+        }
+    }
+}
+
+impl From<LoginRequest> for LoginCommand {
+    fn from(request: LoginRequest) -> Self {
+        match request {
+            LoginRequest::Admin { username, password } => Self::Admin { username, password },
+            LoginRequest::Key { api_key } => Self::Key { api_key },
+        }
+    }
+}
+
+#[derive(Debug, Serialize)]
+#[serde(rename_all = "camelCase")]
+struct SessionData {
+    role: &'static str,
+    expires_at: String,
+}
+
+impl From<&AuthSession> for SessionData {
+    fn from(session: &AuthSession) -> Self {
+        Self {
+            role: match session.subject {
+                SessionSubject::Admin { .. } => "admin",
+                SessionSubject::Key { .. } => "key",
+            },
+            expires_at: session.expires_at.to_rfc3339(),
+        }
+    }
+}
+
+#[derive(Debug, Serialize)]
+struct SessionStatusData {
+    authenticated: bool,
+    session: Option<SessionData>,
+}
+
+#[derive(Debug, Serialize)]
+struct LogoutData {
+    message: &'static str,
+}
+
+pub(crate) fn router<S>() -> Router<S>
+where
+    S: SessionState + Clone + Send + Sync + 'static,
+{
+    Router::new()
+        .route("/api/auth/login", post(login::<S>))
+        .route("/api/auth/status", get(session_status::<S>))
+        .route("/api/auth/logout", post(logout::<S>))
+        .route("/api/auth", any(not_found))
+        .route("/api/auth/{*path}", any(not_found))
+        .method_not_allowed_fallback(method_not_allowed)
+        .layer(middleware::map_response(no_store))
+}
+
+async fn login<S>(
+    State(state): State<S>,
+    ConnectInfo(peer): ConnectInfo<SocketAddr>,
+    headers: HeaderMap,
+    AdminJson(payload): AdminJson<LoginRequest>,
+) -> Result<Response, AdminError>
+where
+    S: SessionState + Send + Sync,
+{
+    let result = state
+        .admin_services()
+        .auth()
+        .login(
+            payload.into(),
+            peer.ip(),
+            session_cookie::value(&headers).as_deref(),
+        )
+        .await
+        .map_err(map_login_error)?;
+    let max_age = result
+        .session
+        .expires_at
+        .signed_duration_since(chrono::Utc::now())
+        .num_seconds()
+        .max(1);
+    let expires = result
+        .session
+        .expires_at
+        .format("%a, %d %b %Y %H:%M:%S GMT");
+    let cookie = format!(
+        "{}={}; {}; Max-Age={max_age}; Expires={expires}",
+        session_cookie::NAME,
+        result.session_id,
+        session_cookie::attributes(&headers)
+    );
+    let mut response = AdminResponse::new(
+        StatusCode::OK,
+        AdminEnvelope::ok(SessionData::from(&result.session)),
+    )
+    .into_response();
+    response.headers_mut().insert(
+        header::SET_COOKIE,
+        HeaderValue::from_str(&cookie).map_err(|_| AdminError::internal())?,
+    );
+    Ok(response)
+}
+
+async fn session_status<S>(
+    State(state): State<S>,
+    headers: HeaderMap,
+) -> Result<impl IntoResponse, AdminError>
+where
+    S: SessionState + Send + Sync,
+{
+    let session = state
+        .admin_services()
+        .auth()
+        .session(session_cookie::value(&headers).as_deref())
+        .await
+        .map_err(map_admin_service_error)?;
+    Ok(AdminResponse::new(
+        StatusCode::OK,
+        AdminEnvelope::ok(SessionStatusData {
+            authenticated: session.is_some(),
+            session: session.as_ref().map(SessionData::from),
+        }),
+    ))
+}
+
+async fn logout<S>(State(state): State<S>, headers: HeaderMap) -> Result<Response, AdminError>
+where
+    S: SessionState + Send + Sync,
+{
+    if let Some(session_id) = session_cookie::value(&headers) {
+        state
+            .admin_services()
+            .auth()
+            .logout(&session_id)
+            .await
+            .map_err(map_admin_service_error)?;
+    }
+    let mut response = AdminResponse::new(
+        StatusCode::OK,
+        AdminEnvelope::ok(LogoutData {
+            message: "Logged out successfully",
+        }),
+    )
+    .into_response();
+    let cookie = format!(
+        "{}=; {}; Max-Age=0",
+        session_cookie::NAME,
+        session_cookie::attributes(&headers)
+    );
+    response.headers_mut().insert(
+        header::SET_COOKIE,
+        HeaderValue::from_str(&cookie).map_err(|_| AdminError::internal())?,
+    );
+    Ok(response)
+}
+
+fn map_login_error(error: LoginError) -> AdminError {
+    match error {
+        LoginError::InvalidCredentials => AdminError::invalid_credentials(),
+        LoginError::TooManyAttempts {
+            retry_after_seconds,
+        } => AdminError::too_many_login_attempts().with_retry_after(retry_after_seconds),
+        LoginError::Unavailable => AdminError::service_unavailable(),
+    }
+}
+
+async fn method_not_allowed() -> AdminError {
+    AdminError::method_not_allowed()
+}
+async fn not_found() -> AdminError {
+    AdminError::not_found("认证接口不存在")
+}
+async fn no_store(mut response: Response) -> Response {
+    response
+        .headers_mut()
+        .insert(header::CACHE_CONTROL, HeaderValue::from_static("no-store"));
+    response
+}

backend/crates/gateway-api/src/key_usage/mod.rs

@@ -0,0 +1,93 @@
+//! 当前 Key 的单页用量 API；只暴露页面需要的只读字段。
+
+use axum::{
+    Router,
+    extract::State,
+    http::{HeaderMap, HeaderValue, StatusCode, header},
+    middleware,
+    response::{IntoResponse, Response},
+    routing::{any, get},
+};
+
+use crate::{
+    admin::{AdminEnvelope, AdminError, AdminQuery, AdminResponse, wire::map_admin_service_error},
+    auth::SessionState,
+    session_cookie,
+};
+
+mod presenter;
+mod query;
+
+pub(crate) fn router<S>() -> Router<S>
+where
+    S: SessionState + Clone + Send + Sync + 'static,
+{
+    Router::new()
+        .route("/api/key-usage/overview", get(overview::<S>))
+        .route("/api/key-usage/records", get(records::<S>))
+        .route("/api/key-usage", any(not_found))
+        .route("/api/key-usage/{*path}", any(not_found))
+        .method_not_allowed_fallback(method_not_allowed)
+        .layer(middleware::map_response(no_store))
+}
+
+async fn overview<S>(
+    State(state): State<S>,
+    headers: HeaderMap,
+    AdminQuery(query): AdminQuery<query::OverviewQuery>,
+) -> Result<impl IntoResponse, AdminError>
+where
+    S: SessionState + Send + Sync,
+{
+    let overview = state
+        .admin_services()
+        .key_usage()
+        .overview(
+            session_cookie::value(&headers).as_deref(),
+            query.into_domain()?,
+        )
+        .await
+        .map_err(map_admin_service_error)?
+        .ok_or_else(AdminError::session_required)?;
+    Ok(AdminResponse::new(
+        StatusCode::OK,
+        AdminEnvelope::ok(presenter::overview(overview)),
+    ))
+}
+
+async fn records<S>(
+    State(state): State<S>,
+    headers: HeaderMap,
+    AdminQuery(query): AdminQuery<query::RecordsQuery>,
+) -> Result<impl IntoResponse, AdminError>
+where
+    S: SessionState + Send + Sync,
+{
+    let records = state
+        .admin_services()
+        .key_usage()
+        .records(
+            session_cookie::value(&headers).as_deref(),
+            query.into_domain()?,
+        )
+        .await
+        .map_err(map_admin_service_error)?
+        .ok_or_else(AdminError::session_required)?;
+    Ok(AdminResponse::new(
+        StatusCode::OK,
+        AdminEnvelope::ok(presenter::records(records)),
+    ))
+}
+
+async fn not_found() -> AdminError {
+    AdminError::not_found("密钥用量接口不存在")
+}
+async fn method_not_allowed() -> AdminError {
+    AdminError::method_not_allowed()
+}
+async fn no_store(mut response: Response) -> Response {
+    response
+        .headers_mut()
+        .insert(header::CACHE_CONTROL, HeaderValue::from_static("no-store"));
+    response
+}

backend/crates/gateway-api/src/key_usage/presenter.rs

@@ -0,0 +1,222 @@
+//! 只投影当前 Key 可见的用量字段，不序列化账号、凭据、上游标识或诊断正文。
+
+use chrono::{DateTime, Utc};
+use gateway_admin::model::{
+    key_usage::{KeyUsageOverview, KeyUsageRecords},
+    observability::{
+        CostCoverage, CurrencyCost, Granularity, OpsError, RequestMetrics, UsageListRecord,
+    },
+};
+use serde::Serialize;
+
+use crate::admin::observability::{
+    BillingView, HealthTimelineView, PageData, TokenDetailsView, billing_view,
+    health_timeline_view, usage_list_token_details,
+};
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+pub(super) struct OverviewView {
+    as_of: DateTime<Utc>,
+    start_time: DateTime<Utc>,
+    end_time: DateTime<Utc>,
+    key: KeyView,
+    summary: MetricsView,
+    trend: Vec<TrendPointView>,
+    health_timeline: HealthTimelineView,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct KeyView {
+    name: String,
+    prefix: String,
+    max_concurrency: u64,
+    requests_per_minute: u64,
+    daily_limit_usd: String,
+    daily_used_usd: String,
+    daily_resets_at: Option<DateTime<Utc>>,
+    weekly_limit_usd: String,
+    weekly_used_usd: String,
+    weekly_resets_at: Option<DateTime<Utc>>,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct MetricsView {
+    requests: u64,
+    input_tokens: u64,
+    output_tokens: u64,
+    cached_tokens: u64,
+    cache_write_tokens: u64,
+    reasoning_tokens: u64,
+    total_tokens: u64,
+    cost_usd: Option<String>,
+    cost_incomplete: bool,
+}
+
+fn metrics(value: &RequestMetrics, costs: &[CurrencyCost], coverage: &CostCoverage) -> MetricsView {
+    MetricsView {
+        requests: value.request_count,
+        input_tokens: value.input_tokens,
+        output_tokens: value.output_tokens,
+        cached_tokens: value.cached_tokens,
+        cache_write_tokens: value.cache_write_tokens,
+        reasoning_tokens: value.reasoning_tokens,
+        total_tokens: value.total_tokens,
+        cost_usd: costs
+            .iter()
+            .find(|cost| cost.currency.eq_ignore_ascii_case("USD"))
+            .map(|cost| cost.amount.as_str().to_owned())
+            .or_else(|| (value.request_count == 0).then(|| "0".to_owned())),
+        cost_incomplete: coverage.partial_count > 0 || coverage.unavailable_count > 0,
+    }
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct TrendPointView {
+    time: DateTime<Utc>,
+    bucket_seconds: u32,
+    #[serde(flatten)]
+    metrics: MetricsView,
+}
+
+pub(super) fn overview(value: KeyUsageOverview) -> OverviewView {
+    let key = value.key;
+    OverviewView {
+        as_of: Utc::now(),
+        start_time: value.overview.range.start,
+        end_time: value.overview.range.end,
+        key: KeyView {
+            name: key.name,
+            prefix: key.prefix,
+            max_concurrency: key.limits.max_concurrency,
+            requests_per_minute: key.limits.requests_per_minute,
+            daily_limit_usd: key.budget.limits.daily_usd.canonical(),
+            daily_used_usd: key.budget.daily_used_usd.canonical(),
+            daily_resets_at: key.budget.daily_resets_at.map(DateTime::from),
+            weekly_limit_usd: key.budget.limits.weekly_usd.canonical(),
+            weekly_used_usd: key.budget.weekly_used_usd.canonical(),
+            weekly_resets_at: key.budget.weekly_resets_at.map(DateTime::from),
+        },
+        summary: metrics(
+            &value.overview.requests,
+            &value.overview.attempts.costs,
+            &value.overview.attempts.cost_coverage,
+        ),
+        trend: value
+            .trend
+            .into_iter()
+            .map(|point| TrendPointView {
+                time: point.bucket_start,
+                bucket_seconds: match point.granularity {
+                    Granularity::FifteenMinutes => 900,
+                    Granularity::Hour => 3600,
+                    Granularity::Day => 86400,
+                },
+                metrics: metrics(&point.metrics, &point.costs, &point.cost_coverage),
+            })
+            .collect(),
+        health_timeline: health_timeline_view(value.health_timeline),
+    }
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+pub(super) struct RecordView {
+    id: String,
+    created_at: DateTime<Utc>,
+    model: Option<String>,
+    route: Option<String>,
+    reasoning_effort: Option<String>,
+    client_transport: Option<String>,
+    upstream_transport: Option<String>,
+    token_details: Option<TokenDetailsView>,
+    billing: Option<BillingView>,
+    latency_ms: Option<u64>,
+    first_token_latency_ms: Option<u64>,
+    latency_details: OutputTimingView,
+    client_ip: Option<String>,
+    user_agent: Option<String>,
+    status: &'static str,
+    status_code: Option<u16>,
+}
+
+// Key 只查看自身请求的输出时间，不包含账号容量、调度等待等管理侧观测。
+#[derive(Default, Serialize)]
+#[serde(rename_all = "camelCase")]
+struct OutputTimingView {
+    #[serde(skip_serializing_if = "Option::is_none")]
+    first_event_ms: Option<u64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    first_reasoning_ms: Option<u64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    first_text_ms: Option<u64>,
+}
+
+fn success_record(value: UsageListRecord) -> RecordView {
+    let token_details = Some(usage_list_token_details(&value));
+    let billing = billing_view(value.billing.as_ref());
+    RecordView {
+        id: value.id,
+        created_at: value.started_at,
+        model: value.requested_model_id,
+        route: Some(value.endpoint),
+        reasoning_effort: value.reasoning_effort,
+        client_transport: Some(value.client_transport),
+        upstream_transport: value.upstream_transport,
+        token_details,
+        billing,
+        latency_ms: value.latency_ms,
+        first_token_latency_ms: value.first_token_ms,
+        latency_details: OutputTimingView {
+            first_event_ms: value.first_event_ms,
+            first_reasoning_ms: value.first_reasoning_ms,
+            first_text_ms: value.first_text_ms,
+        },
+        client_ip: value.client_ip,
+        user_agent: value.user_agent,
+        status: "success",
+        // 成功记录不保存 HTTP 状态，不能用 200 伪造缺失的原始事实。
+        status_code: None,
+    }
+}
+
+fn error_record(value: OpsError) -> RecordView {
+    RecordView {
+        id: value.event_id,
+        created_at: value.occurred_at,
+        model: value.requested_model_id,
+        route: value.endpoint,
+        reasoning_effort: value.reasoning_effort,
+        client_transport: value.client_transport,
+        upstream_transport: value.upstream_transport,
+        token_details: None,
+        billing: None,
+        latency_ms: value.latency_ms,
+        first_token_latency_ms: None,
+        latency_details: OutputTimingView::default(),
+        client_ip: value.client_ip,
+        user_agent: value.user_agent,
+        status: "error",
+        status_code: value.client_status_code,
+    }
+}
+
+pub(super) fn records(value: KeyUsageRecords) -> PageData<RecordView> {
+    match value {
+        KeyUsageRecords::Success(page) => PageData {
+            items: page.items.into_iter().map(success_record).collect(),
+            current_page: page.current_page,
+            page_size: page.page_size,
+            total: page.total,
+        },
+        KeyUsageRecords::Error(page) => PageData {
+            items: page.items.into_iter().map(error_record).collect(),
+            current_page: page.current_page,
+            page_size: page.page_size,
+            total: page.total,
+        },
+    }
+}

backend/crates/gateway-api/src/key_usage/query.rs

@@ -0,0 +1,106 @@
+//! 查询不接受 Key、账号或 Provider 范围；禁止未知字段穿透权限边界。
+
+use axum::http::StatusCode;
+use chrono::{DateTime, Duration};
+use gateway_admin::model::{
+    PageSize,
+    key_usage::{KeyUsageQuery, KeyUsageRecordKind, KeyUsageRecordsQuery},
+    observability::TimeRange,
+};
+use serde::Deserialize;
+
+use crate::admin::AdminError;
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase", deny_unknown_fields)]
+pub(super) struct OverviewQuery {
+    start_time: String,
+    end_time: String,
+    model: Option<String>,
+}
+
+impl OverviewQuery {
+    pub(super) fn into_domain(self) -> Result<KeyUsageQuery, AdminError> {
+        let parse = |value: &str| {
+            DateTime::parse_from_rfc3339(value)
+                .map(|value| value.to_utc())
+                .map_err(|_| {
+                    AdminError::invalid_request(
+                        StatusCode::BAD_REQUEST,
+                        "时间必须使用 RFC3339 格式",
+                    )
+                })
+        };
+        let range = TimeRange::new(parse(&self.start_time)?, parse(&self.end_time)?)
+            .map_err(|_| AdminError::invalid_request(StatusCode::BAD_REQUEST, "时间范围不合法"))?;
+        if range.end - range.start > Duration::days(31) {
+            return Err(AdminError::invalid_request(
+                StatusCode::BAD_REQUEST,
+                "一次最多查询 31 天的用量",
+            ));
+        }
+        let model = self
+            .model
+            .map(|value| value.trim().to_owned())
+            .filter(|value| !value.is_empty());
+        if model
+            .as_ref()
+            .is_some_and(|value| value.len() > 256 || value.chars().any(char::is_control))
+        {
+            return Err(AdminError::invalid_request(
+                StatusCode::BAD_REQUEST,
+                "模型筛选内容不合法",
+            ));
+        }
+        Ok(KeyUsageQuery { range, model })
+    }
+}
+
+#[derive(Default, Deserialize)]
+#[serde(rename_all = "camelCase")]
+enum RecordKind {
+    #[default]
+    Success,
+    Error,
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase", deny_unknown_fields)]
+pub(super) struct RecordsQuery {
+    start_time: String,
+    end_time: String,
+    model: Option<String>,
+    current_page: Option<u32>,
+    page_size: Option<u16>,
+    #[serde(default)]
+    kind: RecordKind,
+}
+
+impl RecordsQuery {
+    pub(super) fn into_domain(self) -> Result<KeyUsageRecordsQuery, AdminError> {
+        let current_page = self.current_page.unwrap_or(1);
+        let page_size = self.page_size.unwrap_or(20);
+        if current_page == 0 || !(1..=100).contains(&page_size) {
+            return Err(AdminError::invalid_request(
+                StatusCode::BAD_REQUEST,
+                "页码必须大于 0，每页数量为 1–100",
+            ));
+        }
+        Ok(KeyUsageRecordsQuery {
+            usage: OverviewQuery {
+                start_time: self.start_time,
+                end_time: self.end_time,
+                model: self.model,
+            }
+            .into_domain()?,
+            kind: match self.kind {
+                RecordKind::Success => KeyUsageRecordKind::Success,
+                RecordKind::Error => KeyUsageRecordKind::Error,
+            },
+            current_page,
+            page_size: PageSize::new(page_size).map_err(|_| {
+                AdminError::invalid_request(StatusCode::BAD_REQUEST, "每页数量不合法")
+            })?,
+        })
+    }
+}

backend/crates/gateway-api/src/lib.rs

@@ -29,8 +29,11 @@ use crate::health::HealthStatus;
 use crate::openai::service::OpenAiService;
 
 pub mod admin;
+pub mod auth;
 mod health;
+mod key_usage;
 pub mod openai;
+mod session_cookie;
 
 /// API-owned HTTP 与静态资源配置。
 #[derive(Debug, Clone, Deserialize, PartialEq, Eq)]
@@ -136,6 +139,8 @@ pub fn initialize(
         .route("/healthz", get(health::healthz))
         .merge(openai::router::router())
         .merge(admin::router::<ApiState>())
+        .merge(auth::router::<ApiState>())
+        .merge(key_usage::router::<ApiState>())
         .fallback_service(ServeDir::new(config.asset_directory).fallback(ServeFile::new(index)));
     if !config.cors_allowed_origins.is_empty() {
         let origins = config
@@ -225,7 +230,7 @@ impl ApiState {
     }
 }
 
-impl admin::AdminSessionState for ApiState {
+impl auth::SessionState for ApiState {
     fn admin_services(&self) -> &AdminServices {
         &self.admin
     }

backend/crates/gateway-api/src/openai/error.rs

@@ -247,6 +247,16 @@ pub const fn gateway_error_contract(
             "server_error",
             "provider_infrastructure_unavailable",
         ),
+        GatewayErrorKind::ConcurrencyQueueFull => (
+            StatusCode::TOO_MANY_REQUESTS,
+            "rate_limit_error",
+            "concurrency_queue_full",
+        ),
+        GatewayErrorKind::ConcurrencyQueueTimeout => (
+            StatusCode::TOO_MANY_REQUESTS,
+            "rate_limit_error",
+            "concurrency_queue_timeout",
+        ),
         GatewayErrorKind::RateLimited => (
             StatusCode::TOO_MANY_REQUESTS,
             "rate_limit_error",

backend/crates/gateway-api/src/openai/responses/response.rs

@@ -77,6 +77,21 @@ impl OpenAiResponsesEncoder {
             return Vec::new();
         };
         self.observe_wire(wire);
+        // WS 客户端只对它无法消费的裸 `error` 帧做投影：codex 的 WS 端点
+        // 会静默忽略缺少 status 且不含内置可重试错误码的 `error` 帧，客户
+        // 端只能空等到 idle 超时。投影成 `response.failed`（消息 JSON 自带
+        // type 字段），codex 将其映射为可重试错误并立即重试。可消费形状
+        // （带非 2xx status 或特殊错误码）原样透传，保持业务语义。
+        if wire.event_type() == Some("error")
+            && !ws_client_consumable_error(wire.data())
+            && let Some(data) = response_failed_sse_data_from_error_event(
+                self.response_snapshot.as_ref(),
+                self.response_id.as_deref(),
+                wire.data(),
+            )
+        {
+            return vec![data.to_string()];
+        }
         vec![wire.data().to_string()]
     }
 
@@ -156,3 +171,29 @@ fn openai_wire(event: &ProviderEvent) -> Option<&ProtocolWireEvent> {
         .wire_event()
         .filter(|wire| wire.protocol() == OPENAI_PROTOCOL)
 }
+
+/// 判断 WS 上游的 `error` 帧是否已经是客户端可直接消费的形状。
+///
+/// codex 的 WS 端点只消费两类 `error` 帧：带非 2xx HTTP status 的包装错误，
+/// 以及连接数上限 / previous_response_not_found 这类内置可重试错误码；其余
+/// 帧（包括 status 为 2xx 的矛盾形状）会被静默忽略。可消费的帧原样透传，
+/// 不可消费的才在 WS 边界投影成 `response.failed`。
+fn ws_client_consumable_error(data: &Value) -> bool {
+    if data
+        .get("error")
+        .and_then(|error| error.get("code"))
+        .and_then(Value::as_str)
+        .is_some_and(|code| {
+            matches!(
+                code,
+                "websocket_connection_limit_reached" | "previous_response_not_found"
+            )
+        })
+    {
+        return true;
+    }
+    data.get("status")
+        .or_else(|| data.get("status_code"))
+        .and_then(Value::as_u64)
+        .is_some_and(|status| !(200..300).contains(&status))
+}

backend/crates/gateway-api/src/session_cookie.rs

@@ -0,0 +1,33 @@
+//! 控制面会话 Cookie 的同源 HTTP / HTTPS 属性与解析。
+
+use axum::http::{HeaderMap, header::ORIGIN};
+use url::Url;
+
+pub(crate) const NAME: &str = "cpr_session";
+
+pub(crate) fn attributes(headers: &HeaderMap) -> &'static str {
+    // HTTPS 反代回源 HTTP 不改变浏览器 Origin。只有单个、严格合法的 HTTP Origin
+    // 才允许省略 Secure；缺失、opaque 或非法来源继续 fail closed。
+    let http_origin = headers.get_all(ORIGIN).iter().count() == 1
+        && headers
+            .get(ORIGIN)
+            .and_then(|value| value.to_str().ok())
+            .is_some_and(|origin| {
+                Url::parse(origin).is_ok_and(|url| {
+                    url.scheme() == "http" && url.origin().ascii_serialization() == origin
+                })
+            });
+    if http_origin {
+        "Path=/; HttpOnly; SameSite=Lax"
+    } else {
+        "Path=/; Secure; HttpOnly; SameSite=Lax"
+    }
+}
+
+pub(crate) fn value(headers: &HeaderMap) -> Option<String> {
+    let cookie = headers.get("cookie")?.to_str().ok()?;
+    cookie.split(';').find_map(|part| {
+        let (candidate, value) = part.trim().split_once('=')?;
+        (candidate == NAME).then(|| value.to_owned())
+    })
+}

backend/crates/gateway-api/tests/admin/account_groups.rs

@@ -9,7 +9,7 @@ use tower::ServiceExt as _;
 
 use super::{AdminTestFixture, AdminTestState, PRIMARY_GROUP_ID};
 
-const SESSION_COOKIE: &str = "cpr_admin_session=valid-session";
+const SESSION_COOKIE: &str = "cpr_session=valid-session";
 
 #[tokio::test]
 async fn list_route_should_keep_camel_case_group_and_page_wire() {

backend/crates/gateway-api/tests/admin/accounts/handlers.rs

@@ -31,7 +31,7 @@ async fn personal_info_requires_admin_and_a_valid_account_query() {
             .uri(format!("/api/admin/accounts/personal-info{query}"))
             .header("x-request-id", "req_personal_info");
         if authenticated {
-            request = request.header(header::COOKIE, "cpr_admin_session=valid-session");
+            request = request.header(header::COOKIE, "cpr_session=valid-session");
         }
         let response = admin::router::<AdminTestState>()
             .with_state(fixture.state())
@@ -73,7 +73,7 @@ async fn quota_forecast_requires_admin_and_valid_account_query() {
             .uri(uri)
             .header("x-request-id", "req_forecast");
         if authenticated {
-            request = request.header(header::COOKIE, "cpr_admin_session=valid-session");
+            request = request.header(header::COOKIE, "cpr_session=valid-session");
         }
         let response = admin::router::<AdminTestState>()
             .with_state(fixture.state())

backend/crates/gateway-api/tests/admin/accounts/mod.rs

@@ -319,31 +319,29 @@ mod batch_update {
     }
 
     #[test]
-    fn batch_update_should_require_enabled_and_reject_unknown_fields() {
+    fn batch_update_model_access_preserves_omitted_settings_and_rejects_unknown_fields() {
+        let request: BatchUpdateAccountsRequest = serde_json::from_value(json!({
+            "accountIds": ["acct_test"], "modelAccess": {"mode":"allowlist","models":["test-luna"]}
+        }))
+        .expect("policy-only update");
+        request.validate().expect("valid update");
+        assert!(request.enabled.is_none());
+        assert!(request.weight.is_none());
+        assert!(request.group_ids.is_none());
+        assert!(request.concurrency_limit.is_none());
         assert!(
             serde_json::from_value::<BatchUpdateAccountsRequest>(json!({
-            "accountIds": ["acct_test"],
-            "concurrencyLimit": null,
-            "weight": 1,
-            "groupIds": []
-            }))
-            .is_err()
-        );
-        assert!(
-            serde_json::from_value::<BatchUpdateAccountsRequest>(json!({
-                "accountIds": ["acct_test"],
-                "enabled": true,
-                "concurrencyLimit": null,
-                "weight": 1,
-                "groupIds": [],
-                "legacy": true
+                "accountIds": ["acct_test"], "enabled": true, "legacy": true
             }))
             .is_err()
         );
+        let empty: BatchUpdateAccountsRequest =
+            serde_json::from_value(json!({"accountIds":["acct_test"]})).expect("empty mutation");
+        assert!(empty.validate().is_err());
     }
 
     #[test]
-    fn batch_update_should_reject_invalid_scheduling_bounds_and_missing_fields() {
+    fn batch_update_should_reject_invalid_scheduling_bounds_and_distinguish_clear_from_preserve() {
         for (concurrency_limit, weight, field) in [
             (json!(0), json!(1), "concurrencyLimit"),
             (json!(4294967296_u64), json!(1), "concurrencyLimit"),
@@ -351,36 +349,20 @@ mod batch_update {
             (json!(null), json!(101), "weight"),
         ] {
             let request: BatchUpdateAccountsRequest = serde_json::from_value(json!({
-                "accountIds": ["acct_test"],
-                "enabled": true,
-                "concurrencyLimit": concurrency_limit,
-                "weight": weight,
-                "groupIds": []
-            }))
-            .expect("deserialize invalid scheduling");
+                "accountIds": ["acct_test"], "concurrencyLimit": concurrency_limit, "weight": weight,
+            })).expect("deserialize bounds");
             assert_eq!(
-                request.validate().expect_err("reject scheduling").field(),
+                request.validate().expect_err("reject bounds").field(),
                 field
             );
         }
-        assert!(
-            serde_json::from_value::<BatchUpdateAccountsRequest>(json!({
-                "accountIds": ["acct_test"],
-                "enabled": true,
-                "weight": 1,
-                "groupIds": []
-            }))
-            .is_err()
-        );
-        assert!(
-            serde_json::from_value::<BatchUpdateAccountsRequest>(json!({
-                "accountIds": ["acct_test"],
-                "enabled": true,
-                "concurrencyLimit": null,
-                "groupIds": []
-            }))
-            .is_err()
-        );
+        let cleared: BatchUpdateAccountsRequest = serde_json::from_value(json!({
+            "accountIds": ["acct_test"], "concurrencyLimit": null
+        }))
+        .expect("clear concurrency override");
+        cleared.validate().expect("valid clear");
+        assert_eq!(cleared.concurrency_limit, Some(None));
+        assert!(cleared.weight.is_none());
     }
 
     #[test]
@@ -407,6 +389,43 @@ mod batch_update {
     }
 }
 
+#[test]
+fn single_update_should_accept_optional_unicode_and_multiline_notes() {
+    use gateway_api::admin::accounts::UpdateAccountRequest;
+    use serde_json::json;
+    for notes in [
+        json!(null),
+        json!(""),
+        json!("团队备用\n第二行\t说明"),
+        json!("备".repeat(500)),
+    ] {
+        let request: UpdateAccountRequest = serde_json::from_value(json!({
+            "accountId": "acct_notes", "enabled": true, "concurrencyLimit": null,
+            "weight": 1, "groupIds": [], "notes": notes
+        }))
+        .unwrap();
+        request.validate().expect("accept bounded notes");
+    }
+}
+
+#[test]
+fn single_update_should_reject_oversized_notes_and_control_characters() {
+    use gateway_api::admin::accounts::UpdateAccountRequest;
+    use serde_json::json;
+    for notes in [
+        "备".repeat(501),
+        "备注\0".to_owned(),
+        "备注\u{001b}".to_owned(),
+    ] {
+        let request: UpdateAccountRequest = serde_json::from_value(json!({
+            "accountId": "acct_notes", "enabled": true, "concurrencyLimit": null,
+            "weight": 1, "groupIds": [], "notes": notes
+        }))
+        .unwrap();
+        assert_eq!(request.validate().unwrap_err().field(), "notes");
+    }
+}
+
 mod response {
     use gateway_api::admin::accounts::AccountUsageView;
     use gateway_api::admin::presenter::format_decimal_currency;
@@ -845,6 +864,9 @@ mod import_settings {
             ("weight", json!(0)),
             ("weight", json!(101)),
             ("groupIds", json!(["invalid-group"])),
+            ("notes", json!("备".repeat(501))),
+            ("notes", json!("备注\u{0000}")),
+            ("notes", json!("备注\u{001b}")),
         ] {
             let mut settings =
                 json!({"enabled": false, "concurrencyLimit": null, "weight": 1, "groupIds": []});
@@ -865,6 +887,25 @@ mod import_settings {
         }
     }
 
+    #[test]
+    fn import_and_oauth_accept_optional_unicode_and_multiline_notes() {
+        for notes in [
+            json!(null),
+            json!(""),
+            json!("团队备用\n下月续费\t"),
+            json!("备".repeat(500)),
+        ] {
+            let settings = json!({"enabled": true, "concurrencyLimit": null, "weight": 1, "groupIds": [], "notes": notes});
+            let import: AccountImportRequest = serde_json::from_value(
+                json!({"provider": "openai", "data": {}, "settings": settings}),
+            )
+            .unwrap();
+            let oauth: CompleteAccountAuthorizationRequest = serde_json::from_value(json!({"provider": "xai", "flowId": "flow-test", "callbackUrl": "code", "settings": settings})).unwrap();
+            import.validate().expect("import notes");
+            oauth.validate().expect("OAuth notes");
+        }
+    }
+
     #[test]
     fn import_settings_require_a_complete_explicit_configuration() {
         let settings =

backend/crates/gateway-api/tests/admin/auth.rs

@@ -1,7 +1,5 @@
 use gateway_admin::model::auth::{LoginCommand, LoginError};
-use gateway_api::admin::auth::{
-    AdminLoginData, AdminLoginRequest, AdminLogoutData, AdminSessionStatusData,
-};
+use gateway_api::auth::LoginRequest;
 use serde_json::json;
 
 use super::AdminTestFixture;
@@ -9,18 +7,26 @@ use super::AdminTestFixture;
 #[test]
 fn login_request_should_deny_unknown_fields_and_redact_password_debug() {
     let password = "admin-password-must-not-leak";
-    let request = serde_json::from_value::<AdminLoginRequest>(json!({
+    let request = serde_json::from_value::<LoginRequest>(json!({
+        "mode": "admin",
         "username": "admin@example.invalid",
         "password": password
     }))
     .expect("deserialize login request");
 
     assert!(!format!("{request:?}").contains(password));
-    let (username, parsed_password) = request.into_parts();
+    let LoginCommand::Admin {
+        username,
+        password: parsed_password,
+    } = request.into()
+    else {
+        panic!("admin login expected")
+    };
     assert_eq!(username.as_deref(), Some("admin@example.invalid"));
     assert_eq!(parsed_password, password);
     assert!(
-        serde_json::from_value::<AdminLoginRequest>(json!({
+        serde_json::from_value::<LoginRequest>(json!({
+        "mode": "admin",
             "password": password,
             "rememberMe": true
         }))
@@ -28,39 +34,27 @@ fn login_request_should_deny_unknown_fields_and_redact_password_debug() {
     );
 }
 
-#[test]
-fn auth_responses_should_keep_stable_wire_shapes() {
-    assert_eq!(
-        serde_json::to_value(AdminLoginData::new("2026-07-18T08:00:00+08:00".to_owned()))
-            .expect("serialize login"),
-        json!({ "expiresAt": "2026-07-18T08:00:00+08:00" })
-    );
-    assert_eq!(
-        serde_json::to_value(AdminSessionStatusData::new(true)).expect("serialize status"),
-        json!({ "authenticated": true })
-    );
-    assert_eq!(
-        serde_json::to_value(AdminLogoutData::new()).expect("serialize logout"),
-        json!({ "message": "Logged out successfully" })
-    );
-}
-
 #[tokio::test]
 async fn default_auth_service_should_initialize_login_validate_and_logout() {
     let fixture = AdminTestFixture::new().await;
     let service = fixture.services.auth();
     let session = service
-        .login(LoginCommand {
-            username: Some("admin_1".to_owned()),
-            password: "strong-admin-password".to_owned(),
-        })
+        .login(
+            LoginCommand::Admin {
+                username: Some("admin_1".to_owned()),
+                password: "strong-admin-password".to_owned(),
+            },
+            std::net::Ipv4Addr::LOCALHOST.into(),
+            None,
+        )
         .await
         .expect("login succeeds");
     assert!(
         service
-            .validate_session(Some(&session.session_id))
+            .session(Some(&session.session_id))
             .await
             .expect("validate session")
+            .is_some()
     );
     assert_eq!(
         service
@@ -76,9 +70,10 @@ async fn default_auth_service_should_initialize_login_validate_and_logout() {
         .expect("logout session");
     assert!(
         !service
-            .validate_session(Some(&session.session_id))
+            .session(Some(&session.session_id))
             .await
             .expect("validate logged-out session")
+            .is_some()
     );
     assert_eq!(fixture.auth.audit_count(), 2);
 }
@@ -124,10 +119,14 @@ async fn audit_failure_should_revoke_new_session_before_returning_it() {
         fixture
             .services
             .auth()
-            .login(LoginCommand {
-                username: None,
-                password: "strong-admin-password".to_owned(),
-            })
+            .login(
+                LoginCommand::Admin {
+                    username: None,
+                    password: "strong-admin-password".to_owned(),
+                },
+                std::net::Ipv4Addr::LOCALHOST.into(),
+                None
+            )
             .await
             .expect_err("audit failure rejects login"),
         LoginError::Unavailable

backend/crates/gateway-api/tests/admin/client_keys.rs

@@ -395,7 +395,7 @@ async fn reveal_route_should_use_query_id_and_no_store_response() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/client-keys/reveal?id=key_1")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_reveal")
                 .body(Body::empty())
                 .expect("reveal request"),
@@ -436,7 +436,7 @@ async fn list_route_should_accept_the_full_nonzero_u16_page_size() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/client-keys?limit=65535")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_client_keys_max_page")
                 .body(Body::empty())
                 .expect("client key list request"),

backend/crates/gateway-api/tests/admin/errors.rs

@@ -3,16 +3,18 @@ use axum::{
     body::{Body, to_bytes},
     http::{Method, Request, StatusCode, header},
 };
-use gateway_api::admin;
+use gateway_api::auth::SessionState;
 use serde_json::{Value, json};
 use tower::ServiceExt as _;
 
 use super::{AdminTestFixture, AdminTestState};
 
-const SESSION_COOKIE: &str = "cpr_admin_session=valid-session";
+const SESSION_COOKIE: &str = "cpr_session=valid-session";
 
 fn app(state: AdminTestState) -> Router {
-    admin::router::<AdminTestState>().with_state(state)
+    crate::openai::api_router_with_admin(state.admin_services().clone()).layer(axum::Extension(
+        axum::extract::ConnectInfo(std::net::SocketAddr::from(([127, 0, 0, 1], 41000))),
+    ))
 }
 
 fn request(method: Method, uri: &str, body: Body) -> Request<Body> {
@@ -44,7 +46,7 @@ async fn malformed_login_json_should_use_the_admin_error_envelope() {
     let fixture = AdminTestFixture::new().await;
     let mut request = request(
         Method::POST,
-        "/api/admin/auth/login",
+        "/api/auth/login",
         Body::from(r#"{"password":"secret""#),
     );
     request
@@ -77,8 +79,10 @@ async fn invalid_login_json_data_should_not_echo_the_submitted_value() {
     let submitted = "password-must-not-leak";
     let mut request = request(
         Method::POST,
-        "/api/admin/auth/login",
-        Body::from(json!({ "password": submitted, "rememberMe": true }).to_string()),
+        "/api/auth/login",
+        Body::from(
+            json!({ "mode": "admin", "password": submitted, "rememberMe": true }).to_string(),
+        ),
     );
     request
         .headers_mut()
@@ -111,8 +115,8 @@ async fn login_without_json_content_type_should_keep_415_with_an_admin_envelope(
     let response = app(fixture.state())
         .oneshot(request(
             Method::POST,
-            "/api/admin/auth/login",
-            Body::from(json!({ "password": "secret" }).to_string()),
+            "/api/auth/login",
+            Body::from(json!({ "mode": "admin", "password": "secret" }).to_string()),
         ))
         .await
         .expect("missing JSON content type response");
@@ -229,11 +233,7 @@ async fn unknown_admin_path_should_not_fall_through_to_the_spa() {
 async fn unsupported_admin_method_should_keep_allow_and_return_an_envelope() {
     let fixture = AdminTestFixture::new().await;
     let response = app(fixture.state())
-        .oneshot(request(
-            Method::POST,
-            "/api/admin/auth/status",
-            Body::empty(),
-        ))
+        .oneshot(request(Method::POST, "/api/auth/status", Body::empty()))
         .await
         .expect("unsupported admin method response");
     let allow = response
@@ -280,8 +280,8 @@ async fn admin_auth_failures_should_use_stable_chinese_contracts() {
         .expect("invalid API key response");
     let mut invalid_login = request(
         Method::POST,
-        "/api/admin/auth/login",
-        Body::from(json!({ "password": "wrong-password" }).to_string()),
+        "/api/auth/login",
+        Body::from(json!({ "mode": "admin", "password": "wrong-password" }).to_string()),
     );
     invalid_login
         .headers_mut()
@@ -300,15 +300,11 @@ async fn admin_auth_failures_should_use_stable_chinese_contracts() {
     assert_eq!(
         actual.map(|(status, _, body)| (status, body["code"].clone(), body["message"].clone())),
         [
-            (
-                StatusCode::UNAUTHORIZED,
-                json!(40101),
-                json!("需要管理员登录")
-            ),
+            (StatusCode::UNAUTHORIZED, json!(40101), json!("需要登录")),
             (
                 StatusCode::UNAUTHORIZED,
                 json!(40102),
-                json!("管理员用户名或密码错误")
+                json!("登录凭据错误")
             ),
             (
                 StatusCode::UNAUTHORIZED,
@@ -473,6 +469,8 @@ mod provider {
         };
         AccountPageItem {
             account: AccountRecord {
+                notes: None,
+                model_access: Default::default(),
                 id: "acct_error_test".to_owned(),
                 provider_kind: ProviderKind::new(provider).unwrap(),
                 groups: Vec::new(),

backend/crates/gateway-api/tests/admin/mod.rs

@@ -10,7 +10,7 @@ use async_trait::async_trait;
 use chrono::{DateTime, Duration, Utc};
 use futures::future::BoxFuture;
 use gateway_admin::{
-    AdminConfig, AdminServices, InitialAdminPassword,
+    AdminConfig, AdminServices, ClientConfig, InitialAdminPassword,
     model::{
         MutationContext, Revision,
         account_groups::{
@@ -24,7 +24,7 @@ use gateway_admin::{
             AccountUpdateResult, AccountUsage, AccountUsageWindowQuery, AccountUsageWindowResult,
             AccountsUpdateResult, BatchUpdateAccounts, DeleteAccounts, UpdateAccount,
         },
-        auth::{AdminAuditEvent, AdminSession},
+        auth::{AdminAuditEvent, AuthSession},
         client_distribution::{
             ClientArchitecture, ClientDownloadPackage, ClientDownloadSource,
             CodexDesktopWindowsDownloads,
@@ -66,10 +66,13 @@ use gateway_admin::{
         },
     },
 };
-use gateway_api::admin::AdminSessionState;
+use gateway_api::auth::SessionState;
 use gateway_core::{
     account::{AccountStatusFacts, CredentialState, ProviderAccountId, QuotaState},
-    engine::probe::{AccountProbe, AccountProbeError, AccountProbeRequest, AccountProbeResult},
+    engine::{
+        execution::{ClientAuthenticationError, ClientKeyVerifier},
+        probe::{AccountProbe, AccountProbeError, AccountProbeRequest, AccountProbeResult},
+    },
     policy::{ClientApiKeyId, RateLimits},
     routing::{ConfigRevision, ProviderKind, PublicModelId, UpstreamModelId},
     runtime::SnapshotControl,
@@ -88,6 +91,8 @@ mod wire;
 
 pub(super) struct AdminTestFixture {
     pub services: AdminServices,
+    pub client_key: Arc<Mutex<Option<ClientKeyRecord>>>,
+    pub observations: Arc<Mutex<MemoryObservations>>,
     pub auth: Arc<MemoryAuthStore>,
     pub settings: Arc<MemorySettingsStore>,
     pub usage_records: Arc<Mutex<Vec<UsageListRecord>>>,
@@ -106,10 +111,26 @@ impl AdminTestFixture {
     }
 
     pub async fn with_system(system: Arc<dyn SystemOperations>) -> Self {
+        Self::with_dependencies(system, None).await
+    }
+
+    pub async fn with_key_verifier(
+        verifier: Arc<dyn ClientKeyVerifier>,
+        system: Arc<dyn SystemOperations>,
+    ) -> Self {
+        Self::with_dependencies(system, Some(verifier)).await
+    }
+
+    async fn with_dependencies(
+        system: Arc<dyn SystemOperations>,
+        verifier: Option<Arc<dyn ClientKeyVerifier>>,
+    ) -> Self {
         let api_key = Arc::new(Mutex::new(None));
         let auth = Arc::new(MemoryAuthStore::new(api_key.clone()));
         let settings = Arc::new(MemorySettingsStore::new(api_key));
-        let client_keys = Arc::new(MemoryClientKeyStore);
+        let client_key = Arc::new(Mutex::new(None));
+        let client_keys = Arc::new(MemoryClientKeyStore(client_key.clone()));
+        let observations = Arc::new(Mutex::new(MemoryObservations::default()));
         let account_groups = Arc::new(MemoryAccountGroupStore::new());
         let usage_records = Arc::new(Mutex::new(Vec::new()));
         let usage_detail = Arc::new(Mutex::new(None));
@@ -120,6 +141,7 @@ impl AdminTestFixture {
         let provider_error = Arc::new(Mutex::new(None));
         let account = Arc::new(Mutex::new(None));
         let unused = Arc::new(UnusedStore {
+            observations: observations.clone(),
             usage_records: Arc::clone(&usage_records),
             usage_detail: Arc::clone(&usage_detail),
             diagnostics: Arc::clone(&diagnostics),
@@ -151,17 +173,24 @@ impl AdminTestFixture {
                 default_username: "admin_1".to_owned(),
                 default_password: InitialAdminPassword::new("strong-admin-password"),
             },
+            ClientConfig::default(),
             stores,
-            providers,
-            Arc::new(NoopSnapshot),
-            (Arc::new(NoopProbe), Arc::new(proxies::SuccessfulProbe)),
-            Arc::new(StaticClientDistribution),
-            system,
+            gateway_admin::AdminRuntimePorts {
+                providers,
+                snapshot: Arc::new(NoopSnapshot),
+                account_probe: Arc::new(NoopProbe),
+                proxy_probe: Arc::new(proxies::SuccessfulProbe),
+                client_distribution: Arc::new(StaticClientDistribution),
+                system,
+                client_key_verifier: verifier.unwrap_or_else(|| Arc::new(UnusedClientKeyVerifier)),
+            },
         )
         .await
         .expect("initialize test admin services");
         Self {
             services: bundle.services(),
+            client_key,
+            observations,
             auth,
             settings,
             usage_records,
@@ -207,15 +236,17 @@ impl ClientDistributionResolver for StaticClientDistribution {
 #[derive(Clone)]
 pub(super) struct AdminTestState(AdminServices);
 
-impl AdminSessionState for AdminTestState {
+impl SessionState for AdminTestState {
     fn admin_services(&self) -> &AdminServices {
         &self.0
     }
 }
 
 pub(super) struct MemoryAuthStore {
+    pub(super) enabled: AtomicBool,
+    pub(super) unavailable: AtomicBool,
     password_hash: Mutex<Option<String>>,
-    sessions: Mutex<BTreeMap<String, AdminSession>>,
+    sessions: Mutex<BTreeMap<String, AuthSession>>,
     audits: Mutex<Vec<AdminAuditEvent>>,
     api_key: Arc<Mutex<Option<AdminApiKey>>>,
     fail_audit: AtomicBool,
@@ -224,6 +255,8 @@ pub(super) struct MemoryAuthStore {
 impl MemoryAuthStore {
     fn new(api_key: Arc<Mutex<Option<AdminApiKey>>>) -> Self {
         Self {
+            enabled: AtomicBool::new(false),
+            unavailable: AtomicBool::new(false),
             password_hash: Mutex::new(None),
             sessions: Mutex::new(BTreeMap::new()),
             audits: Mutex::new(Vec::new()),
@@ -235,8 +268,10 @@ impl MemoryAuthStore {
     pub fn insert_session(&self, session_id: &str) {
         self.sessions.lock().expect("sessions").insert(
             session_id.to_owned(),
-            AdminSession {
-                admin_user_id: "admin_1".to_owned(),
+            AuthSession {
+                subject: gateway_admin::model::auth::SessionSubject::Admin {
+                    admin_user_id: "admin_1".to_owned(),
+                },
                 expires_at: Utc::now() + Duration::hours(1),
             },
         );
@@ -282,7 +317,7 @@ impl AuthStore for MemoryAuthStore {
         Ok(self.api_key.lock().expect("API key").clone())
     }
 
-    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
         Ok(self
             .sessions
             .lock()
@@ -291,11 +326,7 @@ impl AuthStore for MemoryAuthStore {
             .cloned())
     }
 
-    async fn store_session(
-        &self,
-        session_id: &str,
-        session: &AdminSession,
-    ) -> AdminStoreResult<()> {
+    async fn store_session(&self, session_id: &str, session: &AuthSession) -> AdminStoreResult<()> {
         self.sessions
             .lock()
             .expect("sessions")
@@ -303,10 +334,30 @@ impl AuthStore for MemoryAuthStore {
         Ok(())
     }
 
-    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
+    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
         Ok(self.sessions.lock().expect("sessions").remove(session_id))
     }
 
+    async fn client_key_enabled(
+        &self,
+        id: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool> {
+        if self.unavailable.load(Ordering::SeqCst) {
+            return Err(unavailable("key status"));
+        }
+        Ok(id.as_str() == "key-42" && self.enabled.load(Ordering::SeqCst))
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        _: std::net::IpAddr,
+        _: u32,
+        _: u32,
+        _: std::time::Duration,
+    ) -> AdminStoreResult<Option<std::time::Duration>> {
+        Ok(None)
+    }
+
     async fn append_audit_event(&self, event: AdminAuditEvent) -> AdminStoreResult<()> {
         if self.fail_audit.load(Ordering::SeqCst) {
             return Err(unavailable("auth audit"));
@@ -357,6 +408,9 @@ impl SettingsStore for MemorySettingsStore {
             refresh_concurrency: command.refresh_concurrency,
             max_concurrent_per_account: command.max_concurrent_per_account,
             request_interval_ms: command.request_interval_ms,
+            max_waiting_per_key: command.max_waiting_per_key,
+            max_waiting_per_account: command.max_waiting_per_account,
+            concurrency_wait_timeout_seconds: command.concurrency_wait_timeout_seconds,
             rotation_strategy: command.rotation_strategy,
             min_codex_desktop_version: command.min_codex_desktop_version,
             min_codex_cli_version: command.min_codex_cli_version,
@@ -395,7 +449,17 @@ impl SettingsStore for MemorySettingsStore {
     }
 }
 
-pub(super) struct MemoryClientKeyStore;
+pub(super) struct MemoryClientKeyStore(Arc<Mutex<Option<ClientKeyRecord>>>);
+
+#[derive(Default)]
+pub(super) struct MemoryObservations {
+    pub summary: Option<UsageOverview>,
+    pub trend: Option<Vec<RequestMetricPoint>>,
+    pub summaries: Vec<(TimeRange, UsageFilter)>,
+    pub trends: Vec<(TimeRange, UsageFilter)>,
+    pub records: Vec<UsageQuery>,
+    pub errors: Vec<OpsErrorQuery>,
+}
 
 pub(super) const PRIMARY_GROUP_ID: &str = "grp_11111111111111111111111111111111";
 pub(super) const SECONDARY_GROUP_ID: &str = "grp_22222222222222222222222222222222";
@@ -636,6 +700,18 @@ fn mutation(
 
 #[async_trait]
 impl ClientKeyStore for MemoryClientKeyStore {
+    async fn get_client_key(
+        &self,
+        id: &ClientApiKeyId,
+    ) -> AdminStoreResult<Option<ClientKeyRecord>> {
+        Ok(self
+            .0
+            .lock()
+            .expect("client key")
+            .clone()
+            .filter(|key| &key.id == id))
+    }
+
     async fn list_client_keys(&self, _: ClientKeyListQuery) -> AdminStoreResult<ClientKeyPage> {
         Ok(ClientKeyPage {
             config_revision: Revision::new(1).expect("revision"),
@@ -703,6 +779,7 @@ impl ClientKeyStore for MemoryClientKeyStore {
 }
 
 struct UnusedStore {
+    observations: Arc<Mutex<MemoryObservations>>,
     usage_records: Arc<Mutex<Vec<UsageListRecord>>>,
     usage_detail: Arc<Mutex<Option<UsageDetail>>>,
     diagnostics: Arc<Mutex<Vec<DiagnosticObservation>>>,
@@ -712,6 +789,14 @@ struct UnusedStore {
     account: Arc<Mutex<Option<AccountPageItem>>>,
 }
 
+struct UnusedClientKeyVerifier;
+
+impl ClientKeyVerifier for UnusedClientKeyVerifier {
+    fn verify_client_key(&self, _: &str) -> Result<ClientApiKeyId, ClientAuthenticationError> {
+        Err(ClientAuthenticationError::InvalidKey)
+    }
+}
+
 #[async_trait]
 impl AccountStore for UnusedStore {
     async fn list_accounts(
@@ -885,10 +970,12 @@ impl ObservabilityStore for UnusedStore {
 
     async fn usage_trend(
         &self,
-        _: TimeRange,
-        _: UsageFilter,
+        range: TimeRange,
+        filter: UsageFilter,
     ) -> AdminStoreResult<Vec<RequestMetricPoint>> {
-        Err(unavailable("usage trend"))
+        let mut data = self.observations.lock().expect("observations");
+        data.trends.push((range, filter));
+        data.trend.clone().ok_or_else(|| unavailable("usage trend"))
     }
 
     fn usage_calculated_billing_facts(
@@ -902,6 +989,11 @@ impl ObservabilityStore for UnusedStore {
     }
 
     async fn list_usage_records(&self, query: UsageQuery) -> AdminStoreResult<UsagePage> {
+        self.observations
+            .lock()
+            .expect("observations")
+            .records
+            .push(query.clone());
         let items = self.usage_records.lock().expect("usage records").clone();
         Ok(UsagePage {
             current_page: query.current_page,
@@ -919,8 +1011,17 @@ impl ObservabilityStore for UnusedStore {
             .ok_or_else(|| unavailable("usage detail"))
     }
 
-    async fn usage_summary(&self, _: TimeRange, _: UsageFilter) -> AdminStoreResult<UsageOverview> {
-        Err(unavailable("usage summary"))
+    async fn usage_summary(
+        &self,
+        range: TimeRange,
+        filter: UsageFilter,
+    ) -> AdminStoreResult<UsageOverview> {
+        let mut data = self.observations.lock().expect("observations");
+        data.summaries.push((range, filter));
+        data.summary
+            .clone()
+            .map(|summary| UsageOverview { range, ..summary })
+            .ok_or_else(|| unavailable("usage summary"))
     }
 
     async fn usage_diagnostics(
@@ -933,6 +1034,11 @@ impl ObservabilityStore for UnusedStore {
     }
 
     async fn list_ops_errors(&self, query: OpsErrorQuery) -> AdminStoreResult<OpsErrorPage> {
+        self.observations
+            .lock()
+            .expect("observations")
+            .errors
+            .push(query.clone());
         let items = self.ops_errors.lock().expect("ops errors").clone();
         Ok(OpsErrorPage {
             current_page: query.current_page,
@@ -1130,6 +1236,9 @@ fn test_runtime_settings() -> RuntimeSettings {
         refresh_concurrency: 2,
         max_concurrent_per_account: 3,
         request_interval_ms: 50,
+        max_waiting_per_key: 0,
+        max_waiting_per_account: 0,
+        concurrency_wait_timeout_seconds: 30,
         rotation_strategy: RotationStrategy::Smart,
         min_codex_desktop_version: None,
         min_codex_cli_version: None,

backend/crates/gateway-api/tests/admin/observability/response.rs

@@ -245,7 +245,7 @@ async fn dashboard_summary_should_include_quota_exhaustion_in_unavailable_headli
         .oneshot(
             Request::builder()
                 .uri("/api/admin/dashboard/summary")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_dashboard_account_summary")
                 .body(Body::empty())
                 .expect("dashboard request"),
@@ -326,7 +326,7 @@ async fn dashboard_summary_should_use_lifetime_totals_for_card_footers() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/dashboard/summary")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_dashboard_lifetime_totals")
                 .body(Body::empty())
                 .expect("dashboard request"),
@@ -391,7 +391,7 @@ async fn dashboard_summary_should_default_to_current_china_day() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/dashboard/summary")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_dashboard_today_range")
                 .body(Body::empty())
                 .expect("dashboard request"),
@@ -480,7 +480,7 @@ async fn usage_detail_should_keep_attempt_snapshot_contract() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/usage/records/detail?id=req_detail")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_usage_detail_snapshot")
                 .body(Body::empty())
                 .expect("usage detail request"),
@@ -575,7 +575,7 @@ async fn zero_attempt_failure_detail_keeps_missing_upstream_facts_and_preparatio
         .oneshot(
             Request::builder()
                 .uri("/api/admin/usage/records/detail?id=req_early")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_zero_attempt_detail")
                 .body(Body::empty())
                 .unwrap(),
@@ -688,7 +688,7 @@ async fn ops_errors_should_keep_account_label_and_authentication_contract() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/operations/errors")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_ops_errors_snapshot")
                 .body(Body::empty())
                 .expect("ops errors request"),
@@ -781,7 +781,7 @@ async fn diagnostics_should_keep_stable_key_and_display_name_contract() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/usage/insights/diagnostics?dimension=account")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_diagnostics_snapshot")
                 .body(Body::empty())
                 .expect("diagnostics request"),
@@ -913,7 +913,7 @@ async fn usage_route_should_forward_a_bounded_unknown_outcome_filter() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/usage/records?outcome=provider_future_state")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_usage_other_outcome")
                 .body(Body::empty())
                 .expect("usage request"),
@@ -1033,7 +1033,7 @@ async fn usage_route_should_expose_table_facts_without_detail_payload() {
         .oneshot(
             Request::builder()
                 .uri("/api/admin/usage/records")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_usage_endpoint")
                 .body(Body::empty())
                 .expect("usage request"),

backend/crates/gateway-api/tests/admin/proxies.rs

@@ -184,7 +184,7 @@ async fn request(
         .uri(path)
         .method(if body.is_some() { "POST" } else { "GET" });
     if authenticated {
-        builder = builder.header(header::COOKIE, "cpr_admin_session=valid-session");
+        builder = builder.header(header::COOKIE, "cpr_session=valid-session");
     }
     let body = body.map_or_else(Body::empty, |body| Body::from(body.to_string()));
     let response = gateway_api::admin::proxies::router::<AdminTestState>()

backend/crates/gateway-api/tests/admin/settings.rs

@@ -17,7 +17,7 @@ fn request(method: Method, path: &str, body: Option<Value>) -> Request<Body> {
     let mut builder = Request::builder()
         .method(method)
         .uri(path)
-        .header(header::COOKIE, "cpr_admin_session=valid-session")
+        .header(header::COOKIE, "cpr_session=valid-session")
         .header("x-request-id", "req_admin_settings");
     let body = if let Some(value) = body {
         builder = builder.header(header::CONTENT_TYPE, "application/json");
@@ -45,6 +45,9 @@ fn update_body() -> Value {
         "refreshConcurrency": 4,
         "maxConcurrentPerAccount": 5,
         "requestIntervalMs": 25,
+        "maxWaitingPerKey": 0,
+        "maxWaitingPerAccount": 0,
+        "concurrencyWaitTimeoutSeconds": 30,
         "rotationStrategy": "round_robin",
         "minCodexDesktopVersion": "26.825.6671",
         "minCodexCliVersion": "0.40.0",
@@ -104,6 +107,9 @@ fn settings_response_should_cover_the_full_runtime_settings_contract() {
         refresh_concurrency: 4,
         max_concurrent_per_account: 5,
         request_interval_ms: 25,
+        max_waiting_per_key: 0,
+        max_waiting_per_account: 0,
+        concurrency_wait_timeout_seconds: 30,
         rotation_strategy: RotationStrategy::RoundRobin,
         min_codex_desktop_version: Some("26.825.6671".to_owned()),
         min_codex_cli_version: Some("0.40.0".to_owned()),
@@ -128,6 +134,9 @@ fn settings_response_should_cover_the_full_runtime_settings_contract() {
             "refreshConcurrency": 4,
             "maxConcurrentPerAccount": 5,
             "requestIntervalMs": 25,
+            "maxWaitingPerKey": 0,
+            "maxWaitingPerAccount": 0,
+            "concurrencyWaitTimeoutSeconds": 30,
             "rotationStrategy": "round_robin",
             "minCodexDesktopVersion": "26.825.6671",
             "minCodexCliVersion": "0.40.0",
@@ -176,6 +185,9 @@ fn settings_request_and_response_fields_should_stay_in_lockstep() {
         refresh_concurrency: u32::try_from(request.refresh_concurrency).expect("u32"),
         max_concurrent_per_account: u32::try_from(request.max_concurrent_per_account).expect("u32"),
         request_interval_ms: request.request_interval_ms,
+        max_waiting_per_key: 0,
+        max_waiting_per_account: 0,
+        concurrency_wait_timeout_seconds: 30,
         rotation_strategy: RotationStrategy::parse(&request.rotation_strategy)
             .expect("fixture rotation strategy"),
         min_codex_desktop_version: request.min_codex_desktop_version,
@@ -379,7 +391,7 @@ async fn admin_auth_should_accept_a_configured_request_id_header_name() {
     let unlabelled = Request::builder()
         .method(Method::GET)
         .uri("/api/admin/settings")
-        .header(header::COOKIE, "cpr_admin_session=valid-session")
+        .header(header::COOKIE, "cpr_session=valid-session")
         .body(Body::empty())
         .expect("build settings request");
 
@@ -387,3 +399,26 @@ async fn admin_auth_should_accept_a_configured_request_id_header_name() {
 
     assert_eq!(response.status(), StatusCode::OK);
 }
+
+#[test]
+fn concurrency_queue_fields_validate_bounds_and_accept_disabled_queues() {
+    for (field, invalid) in [
+        ("maxWaitingPerKey", 1001),
+        ("maxWaitingPerAccount", 1001),
+        ("concurrencyWaitTimeoutSeconds", 0),
+        ("concurrencyWaitTimeoutSeconds", 121),
+    ] {
+        let mut body = update_body();
+        body[field] = json!(invalid);
+        let request: UpdateRuntimeSettingsRequest = serde_json::from_value(body).unwrap();
+        assert_eq!(request.validate().unwrap_err().field(), field);
+    }
+    let mut body = update_body();
+    body["maxWaitingPerKey"] = json!(0);
+    body["maxWaitingPerAccount"] = json!(1000);
+    body["concurrencyWaitTimeoutSeconds"] = json!(120);
+    serde_json::from_value::<UpdateRuntimeSettingsRequest>(body)
+        .unwrap()
+        .validate()
+        .unwrap();
+}

backend/crates/gateway-api/tests/admin/system.rs

@@ -45,7 +45,7 @@ async fn update_event_stream_should_preserve_event_id_and_download_progress_perc
         .oneshot(
             Request::builder()
                 .uri("/api/admin/system/update/events")
-                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header(header::COOKIE, "cpr_session=valid-session")
                 .header("x-request-id", "req_system_update_events")
                 .body(Body::empty())
                 .expect("update event request"),

backend/crates/gateway-api/tests/admin/wire.rs

@@ -175,12 +175,12 @@ mod response {
                 40001,
             ),
             (
-                AdminError::admin_session_required(),
+                AdminError::session_required(),
                 StatusCode::UNAUTHORIZED,
                 40101,
             ),
             (
-                AdminError::invalid_admin_credentials(),
+                AdminError::invalid_credentials(),
                 StatusCode::UNAUTHORIZED,
                 40102,
             ),

backend/crates/gateway-api/tests/architecture.rs

@@ -54,7 +54,11 @@ fn source_tree_should_match_frozen_machine_manifest() {
         "src/admin/settings.rs",
         "src/admin/system.rs",
         "src/admin/wire.rs",
+        "src/auth.rs",
         "src/health.rs",
+        "src/key_usage/mod.rs",
+        "src/key_usage/query.rs",
+        "src/key_usage/presenter.rs",
         "src/lib.rs",
         "src/openai/auth.rs",
         "src/openai/error.rs",
@@ -76,6 +80,7 @@ fn source_tree_should_match_frozen_machine_manifest() {
         "src/openai/search/http.rs",
         "src/openai/search/mod.rs",
         "src/openai/service.rs",
+        "src/session_cookie.rs",
     ];
     expected.sort_unstable();
 
@@ -104,7 +109,11 @@ fn test_tree_should_match_frozen_rust_mirror() {
         "tests/admin/system.rs",
         "tests/admin/wire.rs",
         "tests/architecture.rs",
+        "tests/auth.rs",
+        "tests/support.rs",
         "tests/health.rs",
+        "tests/key_usage/mod.rs",
+        "tests/key_usage/fixtures.rs",
         "tests/main.rs",
         "tests/openai/auth.rs",
         "tests/openai/error.rs",

backend/crates/gateway-api/tests/auth.rs

@@ -0,0 +1,368 @@
+use std::sync::atomic::Ordering;
+
+use axum::{
+    body::Body,
+    http::{Method, Request, StatusCode, header},
+};
+use serde_json::{Value, json};
+use tower::ServiceExt as _;
+
+use crate::support::{
+    RAW_KEY, auth_app, cookie_request, empty_request, json_request, response_json,
+};
+
+fn session_cookie(response: &axum::response::Response) -> String {
+    response.headers()[header::SET_COOKIE]
+        .to_str()
+        .expect("cookie")
+        .split(';')
+        .next()
+        .expect("cookie pair")
+        .to_owned()
+}
+
+async fn login(
+    app: &axum::Router,
+    body: Value,
+    previous: Option<&str>,
+) -> axum::response::Response {
+    let mut request = json_request(Method::POST, "/api/auth/login", body);
+    if let Some(cookie) = previous {
+        request
+            .headers_mut()
+            .insert(header::COOKIE, cookie.parse().expect("cookie"));
+    }
+    app.clone().oneshot(request).await.expect("login response")
+}
+
+async fn get(app: &axum::Router, path: &str, cookie: &str) -> axum::response::Response {
+    app.clone()
+        .oneshot(cookie_request(Method::GET, path, cookie))
+        .await
+        .expect("GET response")
+}
+
+#[tokio::test]
+async fn unified_login_returns_server_identity_and_rotates_the_previous_session() {
+    let (app, _) = auth_app().await;
+    let key = login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await;
+    assert_eq!(key.status(), StatusCode::OK);
+    let key_cookie = session_cookie(&key);
+    let key_data = response_json(key).await["data"].clone();
+    assert_eq!(key_data["role"], "key");
+    assert!(key_data["expiresAt"].is_string());
+    assert_eq!(key_data.as_object().expect("session object").len(), 2);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &key_cookie).await).await["data"],
+        json!({"authenticated": true, "session": key_data})
+    );
+
+    let admin = login(
+        &app,
+        json!({"mode": "admin", "username": "admin_1", "password": "strong-admin-password"}),
+        Some(&key_cookie),
+    )
+    .await;
+    assert_eq!(admin.status(), StatusCode::OK);
+    let admin_cookie = session_cookie(&admin);
+    let admin_data = response_json(admin).await["data"].clone();
+    assert_eq!(admin_data["role"], "admin");
+    assert!(admin_data["expiresAt"].is_string());
+    assert_eq!(admin_data.as_object().expect("session object").len(), 2);
+    assert_ne!(key_cookie, admin_cookie);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &admin_cookie).await).await["data"],
+        json!({"authenticated": true, "session": admin_data})
+    );
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &key_cookie).await).await["data"],
+        json!({"authenticated": false, "session": null})
+    );
+    assert_eq!(
+        get(&app, "/api/admin/system/version", &admin_cookie)
+            .await
+            .status(),
+        StatusCode::OK
+    );
+
+    let key = login(
+        &app,
+        json!({"mode": "key", "apiKey": RAW_KEY}),
+        Some(&admin_cookie),
+    )
+    .await;
+    assert_eq!(key.status(), StatusCode::OK);
+    let renewed = session_cookie(&key);
+    assert_ne!(renewed, key_cookie);
+    assert_eq!(
+        get(&app, "/api/admin/system/version", &admin_cookie)
+            .await
+            .status(),
+        StatusCode::UNAUTHORIZED
+    );
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &renewed).await).await["data"]["authenticated"],
+        true
+    );
+}
+
+#[tokio::test]
+async fn login_rejects_legacy_type_without_replacing_the_current_session() {
+    let (app, _) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    for body in [
+        json!({"type": "key", "apiKey": RAW_KEY}),
+        json!({"type": "admin", "password": "strong-admin-password"}),
+        json!({"mode": "key", "type": "key", "apiKey": RAW_KEY}),
+        json!({"mode": "admin", "type": "admin", "password": "strong-admin-password"}),
+    ] {
+        let response = login(&app, body, Some(&cookie)).await;
+        assert_eq!(response.status(), StatusCode::UNPROCESSABLE_ENTITY);
+        assert_eq!(response.headers()[header::CACHE_CONTROL], "no-store");
+        assert!(!response.headers().contains_key(header::SET_COOKIE));
+        let error = response_json(response).await.to_string();
+        assert!(!error.contains(RAW_KEY));
+        assert!(!error.contains("strong-admin-password"));
+    }
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"]["session"]["role"],
+        "key"
+    );
+}
+
+#[tokio::test]
+async fn valid_key_session_cannot_read_or_mutate_admin_resources() {
+    let (app, _) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    for path in [
+        "/api/admin/system/version",
+        "/api/admin/accounts",
+        "/api/admin/client-keys",
+        "/api/admin/settings",
+    ] {
+        let response = get(&app, path, &cookie).await;
+        assert_eq!(response.status(), StatusCode::FORBIDDEN, "{path}");
+        assert_eq!(response_json(response).await["code"], 40301);
+    }
+    let mut request = json_request(
+        Method::POST,
+        "/api/admin/settings/admin-api-key/regenerate",
+        json!({}),
+    );
+    request
+        .headers_mut()
+        .insert(header::COOKIE, cookie.parse().unwrap());
+    let response = app.clone().oneshot(request).await.unwrap();
+    assert_eq!(response.status(), StatusCode::FORBIDDEN);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"]["authenticated"],
+        true
+    );
+}
+
+#[tokio::test]
+async fn declared_roles_and_foreign_key_ids_cannot_override_server_authority() {
+    let (app, _) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    for body in [
+        json!({"mode": "admin", "apiKey": RAW_KEY}),
+        json!({"mode": "key", "apiKey": RAW_KEY, "role": "admin"}),
+        json!({"mode": "admin", "password": RAW_KEY, "role": "key"}),
+        json!({"role": "key", "apiKey": RAW_KEY}),
+        json!({"mode": "key", "apiKey": RAW_KEY, "adminUserId": "admin_1"}),
+        json!({"mode": "key", "apiKey": RAW_KEY, "clientKeyId": "other-key"}),
+        json!({"mode": "superadmin", "apiKey": RAW_KEY}),
+        json!({"apiKey": RAW_KEY}),
+    ] {
+        let response = login(&app, body, Some(&cookie)).await;
+        assert_eq!(response.status(), StatusCode::UNPROCESSABLE_ENTITY);
+        assert!(!response_json(response).await.to_string().contains(RAW_KEY));
+    }
+    let invalid = login(
+        &app,
+        json!({"mode": "admin", "password": RAW_KEY}),
+        Some(&cookie),
+    )
+    .await;
+    assert_eq!(invalid.status(), StatusCode::UNAUTHORIZED);
+    assert!(!invalid.headers().contains_key(header::SET_COOKIE));
+    assert_eq!(response_json(invalid).await["code"], 40102);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"]["session"]["role"],
+        "key"
+    );
+}
+
+#[tokio::test]
+async fn data_plane_keys_and_browser_sessions_are_not_interchangeable() {
+    let (app, _) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    assert_eq!(
+        get(&app, "/v1/models", &cookie).await.status(),
+        StatusCode::UNAUTHORIZED
+    );
+    for (name, value) in [
+        (header::AUTHORIZATION.as_str(), format!("Bearer {RAW_KEY}")),
+        ("x-api-key", RAW_KEY.to_owned()),
+    ] {
+        let request = Request::get("/api/auth/status")
+            .header(name, value)
+            .body(Body::empty())
+            .unwrap();
+        let response = app.clone().oneshot(request).await.unwrap();
+        assert_eq!(
+            response_json(response).await["data"]["authenticated"],
+            false
+        );
+    }
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &format!("cpr_session={RAW_KEY}")).await).await
+            ["data"]["authenticated"],
+        false
+    );
+}
+
+#[tokio::test]
+async fn disabled_key_revokes_its_session_and_reenabling_does_not_restore_it() {
+    let (app, store) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    store.enabled.store(false, Ordering::SeqCst);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"],
+        json!({"authenticated": false, "session": null})
+    );
+    assert_eq!(
+        login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None)
+            .await
+            .status(),
+        StatusCode::UNAUTHORIZED
+    );
+    store.enabled.store(true, Ordering::SeqCst);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"]["authenticated"],
+        false
+    );
+}
+
+#[tokio::test]
+async fn key_store_outage_is_unavailable_not_expired_or_authenticated() {
+    let (app, store) = auth_app().await;
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    store.unavailable.store(true, Ordering::SeqCst);
+    for path in ["/api/auth/status", "/api/admin/system/version"] {
+        let response = get(&app, path, &cookie).await;
+        assert_eq!(response.status(), StatusCode::SERVICE_UNAVAILABLE, "{path}");
+        assert!(!response.headers().contains_key(header::SET_COOKIE));
+    }
+    store.unavailable.store(false, Ordering::SeqCst);
+    assert_eq!(
+        response_json(get(&app, "/api/auth/status", &cookie).await).await["data"]["authenticated"],
+        true
+    );
+}
+
+#[tokio::test]
+async fn removed_auth_routes_do_not_fall_back_to_spa_or_accept_old_cookies() {
+    let (app, _) = auth_app().await;
+    for path in ["/api/admin/auth/login", "/api/auth/unknown"] {
+        let response = app
+            .clone()
+            .oneshot(empty_request(Method::GET, path))
+            .await
+            .unwrap();
+        assert_eq!(response.status(), StatusCode::NOT_FOUND);
+        assert_eq!(response_json(response).await["code"], 40401);
+    }
+    let cookie =
+        session_cookie(&login(&app, json!({"mode": "key", "apiKey": RAW_KEY}), None).await);
+    for old_name in ["cpr_admin_session", "cpr_client_session"] {
+        let old_cookie = cookie.replace("cpr_session", old_name);
+        assert_eq!(
+            response_json(get(&app, "/api/auth/status", &old_cookie).await).await["data"]["authenticated"],
+            false
+        );
+    }
+}
+
+#[test]
+fn login_wire_never_prints_either_credential_in_debug() {
+    for body in [
+        json!({"mode": "key", "apiKey": RAW_KEY}),
+        json!({"mode": "admin", "password": RAW_KEY}),
+    ] {
+        let request: gateway_api::auth::LoginRequest = serde_json::from_value(body).unwrap();
+        let debug = format!("{request:?}");
+        assert!(debug.contains("[REDACTED]"));
+        assert!(!debug.contains(RAW_KEY));
+    }
+}
+
+#[tokio::test]
+async fn key_auth_should_issue_restore_and_clear_a_unified_cookie() {
+    let (app, _) = auth_app().await;
+    let login = app
+        .clone()
+        .oneshot(json_request(
+            Method::POST,
+            "/api/auth/login",
+            json!({ "mode": "key", "apiKey": RAW_KEY }),
+        ))
+        .await
+        .expect("client login response");
+
+    assert_eq!(login.status(), StatusCode::OK);
+    assert_eq!(login.headers()[header::CACHE_CONTROL], "no-store");
+    let set_cookie = login.headers()[header::SET_COOKIE]
+        .to_str()
+        .expect("client session cookie")
+        .to_owned();
+    assert!(set_cookie.starts_with("cpr_session=session_"));
+    assert!(set_cookie.contains("Path=/;"));
+    assert!(set_cookie.contains("; Secure; HttpOnly; SameSite=Lax"));
+    assert!(set_cookie.contains("; Max-Age="));
+    assert!(set_cookie.contains("; Expires="));
+    assert!(!set_cookie.contains(RAW_KEY));
+    let cookie = set_cookie
+        .split(';')
+        .next()
+        .expect("cookie pair")
+        .to_owned();
+    let login_body = response_json(login).await;
+    assert_eq!(login_body["data"]["role"], "key");
+    assert!(login_body["data"].get("key").is_none());
+    assert!(!login_body.to_string().contains(RAW_KEY));
+
+    let status = app
+        .clone()
+        .oneshot(cookie_request(Method::GET, "/api/auth/status", &cookie))
+        .await
+        .expect("client status response");
+    assert_eq!(status.status(), StatusCode::OK);
+    let status_body = response_json(status).await;
+    assert_eq!(status_body["data"]["authenticated"], true);
+    assert_eq!(status_body["data"]["session"]["role"], "key");
+
+    let logout = app
+        .clone()
+        .oneshot(cookie_request(Method::POST, "/api/auth/logout", &cookie))
+        .await
+        .expect("client logout response");
+    assert_eq!(logout.status(), StatusCode::OK);
+    let cleared = logout.headers()[header::SET_COOKIE]
+        .to_str()
+        .expect("cleared cookie");
+    assert!(cleared.starts_with("cpr_session=;"));
+    assert!(cleared.contains("Path=/;"));
+    assert!(cleared.contains("Max-Age=0"));
+    let status = app
+        .oneshot(cookie_request(Method::GET, "/api/auth/status", &cookie))
+        .await
+        .expect("logged out");
+    assert_eq!(response_json(status).await["data"]["authenticated"], false);
+}

backend/crates/gateway-api/tests/key_usage/fixtures.rs

@@ -0,0 +1,198 @@
+use crate::{admin::AdminTestFixture, support::key_fixture};
+use chrono::{Duration, Utc};
+use gateway_admin::model::{
+    client_keys::ClientKeyRecord,
+    observability::{
+        AttemptMetrics, CostCoverage, CurrencyCost, DecimalAmount, Granularity, OpsError,
+        RequestMetricPoint, RequestMetrics, TimeRange, UsageBilling, UsageListRecord,
+        UsageOverview,
+    },
+};
+use gateway_core::{
+    engine::budget::{ClientBudgetLimits, ClientBudgetStatus},
+    policy::{ClientApiKeyId, RateLimits},
+};
+
+pub(super) async fn fixture() -> AdminTestFixture {
+    let fixture = key_fixture().await;
+    let now = Utc::now();
+    *fixture.client_key.lock().unwrap() = Some(ClientKeyRecord {
+        id: ClientApiKeyId::new("key-42").unwrap(),
+        name: "Development".to_owned(),
+        label: Some("private-sentinel".to_owned()),
+        prefix: "cpr_live".to_owned(),
+        groups: Vec::new(),
+        provider_kinds: Vec::new(),
+        enabled: true,
+        limits: RateLimits::unlimited(),
+        budget: ClientBudgetStatus {
+            limits: ClientBudgetLimits {
+                daily_usd: "1".parse().unwrap(),
+                weekly_usd: "5".parse().unwrap(),
+            },
+            daily_used_usd: "0.640001".parse().unwrap(),
+            weekly_used_usd: "2.35".parse().unwrap(),
+            daily_resets_at: Some((now + Duration::days(1)).into()),
+            weekly_resets_at: Some((now + Duration::days(7)).into()),
+        },
+        last_used_at: Some(now),
+        created_at: now,
+        updated_at: now,
+    });
+    let metrics = RequestMetrics {
+        request_count: 2,
+        success_count: 1,
+        failure_count: 1,
+        input_tokens: 1000,
+        output_tokens: 100,
+        cached_tokens: 900,
+        reasoning_tokens: 40,
+        total_tokens: 1100,
+        ..Default::default()
+    };
+    let coverage = CostCoverage {
+        calculated_count: 1,
+        unavailable_count: 1,
+        ..Default::default()
+    };
+    let costs = vec![CurrencyCost {
+        currency: "USD".to_owned(),
+        amount: "0.123456".parse::<DecimalAmount>().unwrap(),
+    }];
+    {
+        let mut observations = fixture.observations.lock().unwrap();
+        observations.summary = Some(UsageOverview {
+            range: TimeRange {
+                start: now - Duration::hours(1),
+                end: now,
+            },
+            requests: metrics.clone(),
+            attempts: AttemptMetrics {
+                costs: costs.clone(),
+                cost_coverage: coverage.clone(),
+                ..Default::default()
+            },
+            providers: Vec::new(),
+        });
+        observations.trend = Some(vec![RequestMetricPoint {
+            bucket_start: now,
+            granularity: Granularity::FifteenMinutes,
+            metrics,
+            costs,
+            cost_coverage: coverage,
+        }]);
+    }
+    fixture.usage_records.lock().unwrap().push(usage_record());
+    fixture.ops_errors.lock().unwrap().push(error_record());
+    fixture
+}
+
+fn usage_record() -> UsageListRecord {
+    UsageListRecord {
+        id: "req-visible".to_owned(),
+        endpoint: "/v1/responses".to_owned(),
+        client_transport: "http_sse".to_owned(),
+        requested_model_id: Some("coding".to_owned()),
+        provider_kind: Some("private-sentinel".to_owned()),
+        provider_account_ref: Some("private-sentinel".to_owned()),
+        provider_account_name: Some("private-sentinel".to_owned()),
+        provider_account_email: Some("private-sentinel".to_owned()),
+        provider_account_authentication_kind: Some("private-sentinel".to_owned()),
+        upstream_model_id: Some("private-sentinel".to_owned()),
+        upstream_transport: Some("websocket".to_owned()),
+        service_tier: Some("private-sentinel".to_owned()),
+        input_tokens: Some(1000),
+        output_tokens: Some(100),
+        cached_tokens: Some(900),
+        cache_write_tokens: Some(0),
+        reasoning_tokens: Some(40),
+        image_input_tokens: None,
+        image_output_tokens: None,
+        total_tokens: Some(1100),
+        cost_source: "calculated".to_owned(),
+        cost_amount: Some("0.123456".parse::<DecimalAmount>().unwrap()),
+        cost_currency: Some("USD".to_owned()),
+        billing: Some(UsageBilling::Total {
+            source: "provider_reported".to_owned(),
+            total: CurrencyCost {
+                currency: "USD".to_owned(),
+                amount: "0.123456".parse().unwrap(),
+            },
+        }),
+        transport_decision_wait_ms: None,
+        connect_ms: None,
+        headers_ms: None,
+        first_event_ms: Some(100),
+        first_reasoning_ms: Some(210),
+        first_text_ms: Some(600),
+        first_token_ms: Some(210),
+        provider_processing_ms: None,
+        latency_ms: Some(1250),
+        admission_decision_ms: None,
+        account_selection_wait_ms: None,
+        capacity_used_slots: Some(17),
+        capacity_total_slots: Some(50),
+        client_ip: Some("192.0.2.42".to_owned()),
+        user_agent: Some("key-usage-test/1.0".to_owned()),
+        reasoning_effort: Some("xhigh".to_owned()),
+        reasoning_preset: Some("private-sentinel".to_owned()),
+        subagent_kind: Some("private-sentinel".to_owned()),
+        compact: false,
+        started_at: Utc::now(),
+    }
+}
+
+fn error_record() -> OpsError {
+    OpsError {
+        source: "private-sentinel".to_owned(),
+        event_id: "error-visible".to_owned(),
+        request_id: Some("private-sentinel".to_owned()),
+        attempt_index: None,
+        client_api_key_ref: Some("private-sentinel".to_owned()),
+        component: "private-sentinel".to_owned(),
+        operation: "private-sentinel".to_owned(),
+        protocol: Some("private-sentinel".to_owned()),
+        client_transport: Some("http_sse".to_owned()),
+        requested_model_id: Some("coding".to_owned()),
+        service_tier: Some("private-sentinel".to_owned()),
+        endpoint: Some("/v1/responses".to_owned()),
+        provider_kind: Some("private-sentinel".to_owned()),
+        provider_account_ref: Some("private-sentinel".to_owned()),
+        provider_account_name: Some("private-sentinel".to_owned()),
+        provider_account_email: Some("private-sentinel".to_owned()),
+        provider_account_authentication_kind: Some("private-sentinel".to_owned()),
+        upstream_model_id: Some("private-sentinel".to_owned()),
+        upstream_transport: Some("websocket".to_owned()),
+        failure_kind: "private-sentinel".to_owned(),
+        upstream_send_state: Some("private-sentinel".to_owned()),
+        client_status_code: Some(502),
+        upstream_status_code: Some(503),
+        provider_error_code: Some("private-sentinel".to_owned()),
+        client_response_id: Some("private-sentinel".to_owned()),
+        upstream_request_id: Some("private-sentinel".to_owned()),
+        latency_ms: Some(420),
+        message: "private-sentinel".to_owned(),
+        raw_upstream_error: Some("private-sentinel".to_owned()),
+        client_ip: Some("192.0.2.42".to_owned()),
+        user_agent: Some("key-usage-test/1.0".to_owned()),
+        reasoning_effort: Some("xhigh".to_owned()),
+        reasoning_preset: Some("private-sentinel".to_owned()),
+        request_kind: Some("private-sentinel".to_owned()),
+        subagent_kind: Some("private-sentinel".to_owned()),
+        compact: None,
+        continuation_affinity_hash: Some("private-sentinel".to_owned()),
+        continuation_previous_response_id_hash: Some("private-sentinel".to_owned()),
+        continuation_unavailable_reason: Some("private-sentinel".to_owned()),
+        upstream_connection_id: Some("private-sentinel".to_owned()),
+        upstream_connection_exit_reason: Some("private-sentinel".to_owned()),
+        upstream_connection_age_ms: None,
+        upstream_connection_idle_ms: None,
+        recovery_request_id: Some("private-sentinel".to_owned()),
+        recovered_at: None,
+        recovery_attempt_count: 0,
+        recovery_retry_delay_ms: None,
+        recovery_total_latency_ms: None,
+        occurred_at: Utc::now(),
+        stable_sort_id: "private-sentinel".to_owned(),
+    }
+}

backend/crates/gateway-api/tests/key_usage/mod.rs

@@ -0,0 +1,365 @@
+//! 通过真实路由验证会话隔离、范围收敛和对外字段白名单。
+
+use std::sync::atomic::Ordering;
+
+use axum::{
+    Router,
+    http::{Method, StatusCode, header},
+    response::Response,
+};
+use chrono::{Duration, Utc};
+use gateway_admin::model::observability::{RequestMetrics, china_day_start};
+use serde_json::{Value, json};
+use tower::ServiceExt as _;
+
+use crate::support::{RAW_KEY, cookie_request, empty_request, json_request, response_json};
+
+mod fixtures;
+
+const RANGE: &str = "startTime=2026-09-01T00:00:00Z&endTime=2026-09-02T00:00:00Z";
+
+async fn login(app: &Router, mode: &str) -> String {
+    let body = if mode == "key" {
+        json!({"mode": "key", "apiKey": RAW_KEY})
+    } else {
+        json!({"mode": "admin", "username": "admin_1", "password": "strong-admin-password"})
+    };
+    let response = app
+        .clone()
+        .oneshot(json_request(Method::POST, "/api/auth/login", body))
+        .await
+        .unwrap();
+    assert_eq!(response.status(), StatusCode::OK);
+    response.headers()[header::SET_COOKIE]
+        .to_str()
+        .unwrap()
+        .split(';')
+        .next()
+        .unwrap()
+        .to_owned()
+}
+
+async fn get(app: &Router, resource: &str, suffix: &str, cookie: &str) -> Response {
+    let path = format!("/api/key-usage/{resource}?{RANGE}{suffix}");
+    let response = app
+        .clone()
+        .oneshot(cookie_request(Method::GET, &path, cookie))
+        .await
+        .unwrap();
+    assert_eq!(response.headers()[header::CACHE_CONTROL], "no-store");
+    response
+}
+
+fn assert_fields(value: &Value, expected: &[&str]) {
+    let mut actual: Vec<_> = value
+        .as_object()
+        .unwrap()
+        .keys()
+        .map(String::as_str)
+        .collect();
+    actual.sort_unstable();
+    let mut expected = expected.to_vec();
+    expected.sort_unstable();
+    assert_eq!(actual, expected);
+}
+
+#[tokio::test]
+async fn overview_scopes_every_query_and_projects_only_key_visible_fields() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let cookie = login(&app, "key").await;
+    let response = get(&app, "overview", "&model=%20coding%20", &cookie).await;
+    assert_eq!(response.status(), StatusCode::OK);
+    let data = response_json(response).await["data"].clone();
+    assert_fields(
+        &data,
+        &[
+            "asOf",
+            "startTime",
+            "endTime",
+            "key",
+            "summary",
+            "trend",
+            "healthTimeline",
+        ],
+    );
+    assert_fields(
+        &data["key"],
+        &[
+            "name",
+            "prefix",
+            "maxConcurrency",
+            "requestsPerMinute",
+            "dailyLimitUsd",
+            "dailyUsedUsd",
+            "dailyResetsAt",
+            "weeklyLimitUsd",
+            "weeklyUsedUsd",
+            "weeklyResetsAt",
+        ],
+    );
+    assert_eq!(data["key"]["dailyUsedUsd"], "0.640001");
+    assert_eq!(data["key"]["maxConcurrency"], 0);
+    assert_eq!(data["summary"]["totalTokens"], 1100);
+    assert_eq!(data["summary"]["reasoningTokens"], 40);
+    assert_eq!(data["trend"][0]["reasoningTokens"], 40);
+    assert_eq!(data["summary"]["costUsd"], "0.123456");
+    assert_eq!(data["summary"]["costIncomplete"], true);
+    assert_eq!(data["trend"][0]["bucketSeconds"], 900);
+    assert_eq!(
+        data["healthTimeline"]["points"].as_array().unwrap().len(),
+        96
+    );
+    assert!(!data.to_string().contains("private-sentinel"));
+    let observations = fixture.observations.lock().unwrap();
+    assert_eq!(observations.summaries.len(), 1);
+    assert_eq!(observations.trends.len(), 2);
+    for (_, filter) in observations
+        .summaries
+        .iter()
+        .chain(observations.trends.iter())
+    {
+        assert_eq!(filter.client_api_key_ref.as_deref(), Some("key-42"));
+        assert!(filter.provider_account_ref.is_none());
+        assert!(filter.provider_kind.is_none());
+    }
+    assert_eq!(observations.summaries[0].1.model.as_deref(), Some("coding"));
+    let health = observations
+        .trends
+        .iter()
+        .find(|(_, filter)| filter.model.is_none())
+        .unwrap();
+    assert_eq!(health.0.start, china_day_start(health.0.end));
+    assert!((Utc::now() - health.0.end) < Duration::seconds(5));
+}
+
+#[tokio::test]
+async fn records_keep_pagination_and_hide_admin_and_upstream_data() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let cookie = login(&app, "key").await;
+    for kind in ["success", "error"] {
+        let response = get(
+            &app,
+            "records",
+            &format!("&kind={kind}&model=coding&currentPage=3&pageSize=10"),
+            &cookie,
+        )
+        .await;
+        assert_eq!(response.status(), StatusCode::OK);
+        let data = response_json(response).await["data"].clone();
+        assert_eq!(data["currentPage"], 3);
+        assert_eq!(data["pageSize"], 10);
+        assert_eq!(data["total"], 1);
+        let record = &data["items"][0];
+        assert_fields(
+            record,
+            &[
+                "id",
+                "createdAt",
+                "model",
+                "route",
+                "reasoningEffort",
+                "clientTransport",
+                "upstreamTransport",
+                "tokenDetails",
+                "billing",
+                "latencyMs",
+                "firstTokenLatencyMs",
+                "latencyDetails",
+                "clientIp",
+                "userAgent",
+                "status",
+                "statusCode",
+            ],
+        );
+        assert_eq!(record["status"], kind);
+        assert_eq!(record["model"], "coding");
+        assert_eq!(record["reasoningEffort"], "xhigh");
+        assert_eq!(record["clientTransport"], "http_sse");
+        assert_eq!(record["upstreamTransport"], "websocket");
+        assert_eq!(record["clientIp"], "192.0.2.42");
+        assert_eq!(record["userAgent"], "key-usage-test/1.0");
+        assert!(!data.to_string().contains("private-sentinel"));
+        if kind == "success" {
+            assert_eq!(record["billing"]["totalAmountDisplay"], "$0.1235");
+            assert_eq!(record["tokenDetails"]["reasoningTokens"], 40);
+            assert_eq!(record["tokenDetails"]["totalTokens"], 1100);
+            assert_eq!(record["firstTokenLatencyMs"], 210);
+            assert_fields(
+                &record["latencyDetails"],
+                &["firstEventMs", "firstReasoningMs", "firstTextMs"],
+            );
+            assert_eq!(record["latencyDetails"]["firstEventMs"], 100);
+            assert!(record["statusCode"].is_null());
+        } else {
+            assert_eq!(record["statusCode"], 502);
+            assert!(record["billing"].is_null());
+            assert!(record["tokenDetails"].is_null());
+            assert_eq!(record["latencyDetails"], json!({}));
+        }
+    }
+    let data = fixture.observations.lock().unwrap();
+    assert_eq!(
+        data.records[0].filter.client_api_key_ref.as_deref(),
+        Some("key-42")
+    );
+    assert_eq!(
+        data.errors[0].filter.client_api_key_ref.as_deref(),
+        Some("key-42")
+    );
+    assert_eq!(data.records[0].filter.model.as_deref(), Some("coding"));
+    assert_eq!(data.errors[0].filter.model.as_deref(), Some("coding"));
+}
+
+#[tokio::test]
+async fn missing_admin_and_revoked_sessions_cannot_read_key_usage() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let admin = login(&app, "admin").await;
+    let key = login(&app, "key").await;
+    for resource in ["overview", "records"] {
+        assert_eq!(
+            get(&app, resource, "", "").await.status(),
+            StatusCode::UNAUTHORIZED
+        );
+        assert_eq!(
+            get(&app, resource, "", &admin).await.status(),
+            StatusCode::FORBIDDEN
+        );
+        fixture.auth.enabled.store(false, Ordering::SeqCst);
+        let response = get(&app, resource, "", &key).await;
+        assert_eq!(response.status(), StatusCode::UNAUTHORIZED);
+        assert_eq!(response_json(response).await["code"], 40101);
+        fixture.auth.enabled.store(true, Ordering::SeqCst);
+    }
+    assert!(fixture.observations.lock().unwrap().records.is_empty());
+    assert!(fixture.observations.lock().unwrap().summaries.is_empty());
+}
+
+#[tokio::test]
+async fn unknown_scope_fields_and_unbounded_queries_are_rejected() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let cookie = login(&app, "key").await;
+    for resource in ["overview", "records"] {
+        for extra in [
+            "&clientApiKeyRef=other",
+            "&keyId=other",
+            "&account=other",
+            "&provider=openai",
+            "&model=%00",
+            "&startTime=duplicate",
+        ] {
+            assert!(
+                get(&app, resource, extra, &cookie)
+                    .await
+                    .status()
+                    .is_client_error()
+            );
+        }
+    }
+    for extra in [
+        "&currentPage=0",
+        "&pageSize=0",
+        "&pageSize=101",
+        "&kind=all",
+        "&currentPage=-1",
+    ] {
+        assert!(
+            get(&app, "records", extra, &cookie)
+                .await
+                .status()
+                .is_client_error()
+        );
+    }
+    for range in [
+        "startTime=invalid&endTime=invalid",
+        "startTime=2026-01-01T00:00:00Z&endTime=2026-09-02T00:00:00Z",
+        "startTime=2026-09-02T00:00:00Z&endTime=2026-09-01T00:00:00Z",
+    ] {
+        let response = app
+            .clone()
+            .oneshot(cookie_request(
+                Method::GET,
+                &format!("/api/key-usage/overview?{range}"),
+                &cookie,
+            ))
+            .await
+            .unwrap();
+        assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+    }
+    assert!(fixture.observations.lock().unwrap().summaries.is_empty());
+    assert!(fixture.observations.lock().unwrap().records.is_empty());
+}
+
+#[tokio::test]
+async fn empty_usage_has_zero_cost_but_unknown_pricing_stays_unknown() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let cookie = login(&app, "key").await;
+    for requests in [0, 2] {
+        {
+            let mut data = fixture.observations.lock().unwrap();
+            let summary = data.summary.as_mut().unwrap();
+            summary.requests = RequestMetrics {
+                request_count: requests,
+                ..Default::default()
+            };
+            summary.attempts.costs.clear();
+            summary.attempts.cost_coverage = Default::default();
+            summary.attempts.cost_coverage.unavailable_count = requests;
+            data.trend = Some(Vec::new());
+        }
+        let data = response_json(get(&app, "overview", "", &cookie).await).await["data"].clone();
+        assert_eq!(
+            data["summary"]["costUsd"],
+            if requests == 0 {
+                json!("0")
+            } else {
+                Value::Null
+            }
+        );
+        assert_eq!(data["summary"]["costIncomplete"], requests > 0);
+    }
+}
+
+#[tokio::test]
+async fn overview_does_not_mask_missing_keys_or_unavailable_observations() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services.clone());
+    let cookie = login(&app, "key").await;
+    fixture.observations.lock().unwrap().summary = None;
+    assert_eq!(
+        get(&app, "overview", "", &cookie).await.status(),
+        StatusCode::SERVICE_UNAVAILABLE
+    );
+    *fixture.client_key.lock().unwrap() = None;
+    assert_eq!(
+        get(&app, "overview", "", &cookie).await.status(),
+        StatusCode::UNAUTHORIZED
+    );
+}
+
+#[tokio::test]
+async fn namespace_errors_remain_json_and_uncacheable() {
+    let fixture = fixtures::fixture().await;
+    let app = crate::openai::api_router_with_admin(fixture.services);
+    for (method, path, status) in [
+        (Method::GET, "/api/key-usage/missing", StatusCode::NOT_FOUND),
+        (
+            Method::POST,
+            "/api/key-usage/overview",
+            StatusCode::METHOD_NOT_ALLOWED,
+        ),
+    ] {
+        let response = app
+            .clone()
+            .oneshot(empty_request(method, path))
+            .await
+            .unwrap();
+        assert_eq!(response.status(), status);
+        assert_eq!(response.headers()[header::CACHE_CONTROL], "no-store");
+        assert!(response_json(response).await["code"].is_number());
+    }
+}

backend/crates/gateway-api/tests/main.rs

@@ -1,4 +1,7 @@
 mod admin;
 mod architecture;
+mod auth;
 mod health;
+mod key_usage;
 mod openai;
+mod support;

backend/crates/gateway-api/tests/openai/error.rs

@@ -110,6 +110,25 @@ fn rate_limited_error_should_map_to_openai_retryable_status() {
     );
 }
 
+#[test]
+fn queue_rejections_keep_distinct_retryable_http_codes() {
+    for (kind, code) in [
+        (
+            GatewayErrorKind::ConcurrencyQueueFull,
+            "concurrency_queue_full",
+        ),
+        (
+            GatewayErrorKind::ConcurrencyQueueTimeout,
+            "concurrency_queue_timeout",
+        ),
+    ] {
+        assert_eq!(
+            gateway_error_contract(kind),
+            (StatusCode::TOO_MANY_REQUESTS, "rate_limit_error", code),
+        );
+    }
+}
+
 #[test]
 fn upstream_unavailable_error_should_map_to_bad_gateway() {
     assert_eq!(

backend/crates/gateway-api/tests/openai/mod.rs

@@ -49,6 +49,32 @@ pub(super) async fn api_router(execution: Arc<dyn ExecutionService>) -> axum::Ro
     api_router_with_origins(execution, Vec::new()).await
 }
 
+pub(super) fn api_router_with_admin(admin: gateway_admin::AdminServices) -> axum::Router {
+    gateway_api::initialize(
+        gateway_api::ApiConfig {
+            asset_directory: std::env::temp_dir(),
+            cors_allowed_origins: Vec::new(),
+            request_timeout_seconds: None,
+            request_id_header: "x-request-id".to_owned(),
+        },
+        Arc::new(DefaultExecutionService::new(
+            RuntimeSnapshotHandle::new(snapshot("unused-client-route-key", "openai")),
+            Arc::new(UnusedExecutionStore),
+            ProviderRegistry::default(),
+            Arc::new(UnusedAdmissions),
+            Arc::new(UnusedCircuits),
+            Arc::new(UnusedContinuation),
+            Arc::new(IgnoredClientApiKeyUsage),
+        )),
+        admin,
+        Vec::new(),
+        Arc::new(EmptyWorkerHealth),
+        Arc::new(TestLifecycle::default()),
+    )
+    .expect("API bundle")
+    .router()
+}
+
 pub(super) async fn api_router_with_worker_health(
     execution: Arc<dyn ExecutionService>,
     worker_health: Arc<dyn WorkerHealthSource>,
@@ -266,6 +292,14 @@ impl ExecutionStore for UnusedExecutionStore {
 struct UnusedAdmissions;
 
 impl ClientAdmissionPort for UnusedAdmissions {
+    fn abandon(
+        &self,
+        key: &gateway_core::policy::ClientApiKeyId,
+        request: &gateway_core::engine::ModelRequestId,
+    ) {
+        let _ = futures::FutureExt::now_or_never(self.release(key, request));
+    }
+
     fn admit(
         &self,
         _: ClientAdmissionRequest,

backend/crates/gateway-api/tests/openai/responses/mod.rs

@@ -666,6 +666,87 @@ fn transparent_encoder_should_use_identical_json_for_sse_and_websocket() {
     );
 }
 
+#[test]
+fn transparent_encoder_should_translate_error_wire_to_response_failed_for_websocket() {
+    // codex 的 WS 端点只消费带 status 的包装错误帧；上游缺少 status 的裸
+    // `error` 帧会被静默忽略，客户端只能空等到 idle 超时。WS 边界与 SSE
+    // 边界一致投影成 `response.failed`，codex 将其映射为可重试错误并立即
+    // 重试，而不是把失败原因拖到流 EOF。
+    let response_id = "resp_ws_error";
+    let started = openai_wire_event(
+        Vec::new(),
+        "response.created",
+        json!({
+            "type": "response.created",
+            "response": {"id": response_id, "status": "in_progress"}
+        }),
+    );
+    let error = openai_wire_event(
+        Vec::new(),
+        "error",
+        json!({
+            "type": "error",
+            "error": {
+                "type": "service_unavailable_error",
+                "code": "server_is_overloaded",
+                "message": "Our servers are currently overloaded. Please try again later."
+            },
+            "sequence_number": 2
+        }),
+    );
+
+    let mut websocket_encoder = OpenAiResponsesEncoder::new();
+    websocket_encoder.push_websocket(&started);
+    let messages = websocket_encoder.push_websocket(&error);
+
+    assert_eq!(messages.len(), 1);
+    assert!(websocket_encoder.has_wire_failure());
+    let payload: Value = serde_json::from_str(&messages[0]).expect("response.failed is valid JSON");
+    assert_eq!(payload["type"], "response.failed");
+    assert_eq!(payload["response"]["id"], response_id);
+    assert_eq!(payload["response"]["status"], "failed");
+    assert_eq!(payload["response"]["error"]["code"], "server_is_overloaded");
+
+    // 与 SSE 边界投影到同一份 data 负载，两条客户端通道行为一致。
+    let mut sse_encoder = OpenAiResponsesEncoder::new();
+    sse_encoder.push_sse(&started);
+    let frames = sse_encoder.push_sse(&error);
+    let sse_text = String::from_utf8(
+        frames
+            .iter()
+            .flat_map(|frame| frame.as_ref().iter().copied())
+            .collect(),
+    )
+    .expect("SSE is UTF-8");
+    let sse_data = parse_sse_events(&sse_text)
+        .expect("SSE parses")
+        .into_iter()
+        .map(|event| event.data)
+        .collect::<Vec<_>>();
+    assert_eq!(sse_data, messages);
+
+    // 带非 2xx status 的包装错误帧是 codex 可直接消费的形状，必须原样透传。
+    let mut passthrough_encoder = OpenAiResponsesEncoder::new();
+    passthrough_encoder.push_websocket(&started);
+    let wrapped = openai_wire_event(
+        Vec::new(),
+        "error",
+        json!({
+            "type": "error",
+            "status": 429,
+            "headers": {"x-request-id": "req_raw"},
+            "error": {"type": "custom_type", "code": "custom_code", "message": "raw marker"},
+            "future": {"keep": true}
+        }),
+    );
+    let wrapped_messages = passthrough_encoder.push_websocket(&wrapped);
+    assert_eq!(
+        wrapped_messages,
+        vec![wrapped.wire_event().expect("wire").data().to_string()]
+    );
+    assert!(passthrough_encoder.has_wire_failure());
+}
+
 #[test]
 fn transparent_encoder_should_follow_wire_when_canonical_identity_changes() {
     let mut encoder = OpenAiResponsesEncoder::new();

backend/crates/gateway-api/tests/openai/responses/websocket/mod.rs

@@ -914,6 +914,14 @@ impl ExecutionStore for SettlementPorts {
 }
 
 impl ClientAdmissionPort for SettlementPorts {
+    fn abandon(
+        &self,
+        key: &gateway_core::policy::ClientApiKeyId,
+        request: &gateway_core::engine::ModelRequestId,
+    ) {
+        let _ = futures::FutureExt::now_or_never(self.release(key, request));
+    }
+
     fn admit(
         &self,
         _: ClientAdmissionRequest,

backend/crates/gateway-api/tests/openai/router.rs

@@ -29,7 +29,11 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
             true,
         ),
     ] {
-        let app = api_router_with_origins(ModelsExecution::new(), Vec::new()).await;
+        let app = api_router_with_origins(ModelsExecution::new(), Vec::new())
+            .await
+            .layer(axum::Extension(axum::extract::ConnectInfo(
+                std::net::SocketAddr::from(([127, 0, 0, 1], 41000)),
+            )));
         let request = |path: &str| {
             let mut builder = Request::post(path)
                 .header("content-type", "application/json")
@@ -43,9 +47,9 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
         let response = app
             .clone()
             .oneshot(
-                request("/api/admin/auth/login")
+                request("/api/auth/login")
                     .body(Body::from(
-                        json!({"username": "admin_1", "password": "strong-admin-password"})
+                        json!({"mode": "admin", "username": "admin_1", "password": "strong-admin-password"})
                             .to_string(),
                     ))
                     .unwrap(),
@@ -64,7 +68,7 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
         let response = app
             .clone()
             .oneshot(
-                Request::get("/api/admin/auth/status")
+                Request::get("/api/auth/status")
                     .header("cookie", &session)
                     .body(Body::empty())
                     .unwrap(),
@@ -78,7 +82,7 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
         let response = app
             .clone()
             .oneshot(
-                request("/api/admin/auth/logout")
+                request("/api/auth/logout")
                     .header("cookie", &session)
                     .body(Body::empty())
                     .unwrap(),
@@ -87,7 +91,7 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
             .unwrap();
         assert_eq!(response.status(), StatusCode::OK);
         let cookie = response.headers()["set-cookie"].to_str().unwrap();
-        assert!(cookie.starts_with("cpr_admin_session=;"));
+        assert!(cookie.starts_with("cpr_session=;"));
         let attrs: Vec<_> = cookie.split(';').map(str::trim).collect();
         assert_eq!(attrs.contains(&"Secure"), secure);
         for attribute in ["Path=/", "HttpOnly", "SameSite=Lax", "Max-Age=0"] {
@@ -96,7 +100,7 @@ async fn browser_origin_controls_http_admin_sessions_without_configuration() {
 
         let response = app
             .oneshot(
-                Request::get("/api/admin/auth/status")
+                Request::get("/api/auth/status")
                     .header("cookie", &session)
                     .body(Body::empty())
                     .unwrap(),

backend/crates/gateway-api/tests/support.rs

@@ -0,0 +1,146 @@
+use std::{
+    net::{IpAddr, Ipv4Addr, SocketAddr},
+    sync::{Arc, atomic::Ordering},
+};
+
+use async_trait::async_trait;
+use axum::{
+    body::{Body, to_bytes},
+    extract::ConnectInfo,
+    http::{Method, Request, header},
+};
+use gateway_admin::{
+    model::system::{
+        SystemOperationAccepted, SystemUpdateDetail, SystemUpdateStatus, SystemVersion,
+    },
+    ports::system::{SystemOperationError, SystemOperations, SystemUpdateEventStream},
+};
+use gateway_core::{
+    engine::execution::{ClientAuthenticationError, ClientKeyVerifier},
+    policy::ClientApiKeyId,
+};
+use serde_json::Value;
+
+pub(crate) const RAW_KEY: &str = "cpr_live_candidate_must_not_leak";
+
+pub(crate) async fn auth_app() -> (axum::Router, Arc<crate::admin::MemoryAuthStore>) {
+    let fixture = key_fixture().await;
+    (
+        crate::openai::api_router_with_admin(fixture.services),
+        fixture.auth,
+    )
+}
+
+pub(crate) async fn key_fixture() -> crate::admin::AdminTestFixture {
+    let fixture = crate::admin::AdminTestFixture::with_key_verifier(
+        Arc::new(AcceptingVerifier {
+            key_id: ClientApiKeyId::new("key-42").expect("key ID"),
+        }),
+        Arc::new(VersionSystem),
+    )
+    .await;
+    fixture.auth.enabled.store(true, Ordering::SeqCst);
+    fixture
+}
+
+pub(crate) fn json_request(method: Method, uri: &str, body: Value) -> Request<Body> {
+    let mut request = Request::builder()
+        .method(method)
+        .uri(uri)
+        .header(header::CONTENT_TYPE, "application/json")
+        .header(header::ORIGIN, "https://console.example.test")
+        .body(Body::from(body.to_string()))
+        .expect("client JSON request");
+    request.extensions_mut().insert(ConnectInfo(SocketAddr::new(
+        IpAddr::V4(Ipv4Addr::new(203, 0, 113, 10)),
+        41_000,
+    )));
+    request
+}
+
+pub(crate) fn empty_request(method: Method, uri: &str) -> Request<Body> {
+    Request::builder()
+        .method(method)
+        .uri(uri)
+        .body(Body::empty())
+        .expect("client empty request")
+}
+
+pub(crate) fn cookie_request(method: Method, uri: &str, cookie: &str) -> Request<Body> {
+    Request::builder()
+        .method(method)
+        .uri(uri)
+        .header(header::COOKIE, cookie)
+        .body(Body::empty())
+        .expect("client cookie request")
+}
+
+pub(crate) async fn response_json(response: axum::response::Response) -> Value {
+    let body = to_bytes(response.into_body(), 64 * 1024)
+        .await
+        .expect("read client response body");
+    serde_json::from_slice(&body).expect("parse client response JSON")
+}
+
+struct VersionSystem;
+
+#[async_trait]
+impl SystemOperations for VersionSystem {
+    async fn version(&self) -> Result<SystemVersion, SystemOperationError> {
+        Ok(SystemVersion {
+            version: "3.7.0".to_owned(),
+            git_sha: "internal-revision".to_owned(),
+            build_time: "internal-build-time".to_owned(),
+            deployment_mode: "binary".to_owned(),
+            update_channel: "release".to_owned(),
+            latest_version: "3.8.0".to_owned(),
+            has_update: true,
+            update_cached: true,
+            update_warning: Some("internal-update-warning".to_owned()),
+        })
+    }
+
+    async fn update_detail(&self, _: bool) -> Result<SystemUpdateDetail, SystemOperationError> {
+        unreachable!("client route must not request update details")
+    }
+
+    fn update_events(&self) -> SystemUpdateEventStream {
+        unreachable!("client route must not subscribe to updates")
+    }
+
+    async fn perform_update(
+        &self,
+        _: Option<String>,
+    ) -> Result<SystemOperationAccepted, SystemOperationError> {
+        unreachable!("client route must not perform updates")
+    }
+
+    async fn update_status(&self) -> Result<SystemUpdateStatus, SystemOperationError> {
+        unreachable!("client route must not request update status")
+    }
+
+    async fn rollback(&self) -> Result<SystemOperationAccepted, SystemOperationError> {
+        unreachable!("client route must not roll back")
+    }
+
+    async fn restart(&self) -> Result<SystemOperationAccepted, SystemOperationError> {
+        unreachable!("client route must not restart")
+    }
+}
+
+struct AcceptingVerifier {
+    key_id: ClientApiKeyId,
+}
+
+impl ClientKeyVerifier for AcceptingVerifier {
+    fn verify_client_key(
+        &self,
+        candidate: &str,
+    ) -> Result<ClientApiKeyId, ClientAuthenticationError> {
+        if candidate == RAW_KEY {
+            Ok(self.key_id.clone())
+        } else {
+            Err(ClientAuthenticationError::InvalidKey)
+        }
+    }
+}

backend/crates/gateway-core/src/account/mod.rs

@@ -2,6 +2,11 @@
 
 mod error;
 mod model;
+mod model_access;
+pub use model_access::{
+    AccountModelAccess, AccountModelAccessMode, InvalidAccountModelAccess,
+    MAX_ACCOUNT_ACCESS_MODELS,
+};
 mod proxy;
 pub use proxy::{InvalidOutboundProxy, OutboundProxy};
 pub mod scope;

backend/crates/gateway-core/src/account/model.rs

@@ -669,6 +669,7 @@ pub struct ProviderAccount {
     enabled: bool,
     concurrency_limit: Option<AccountConcurrencyLimit>,
     weight: AccountWeight,
+    model_access: super::AccountModelAccess,
     credential_state: CredentialState,
     quota: QuotaState,
     last_error_reason: Option<AccountErrorReason>,
@@ -704,6 +705,7 @@ impl ProviderAccount {
             enabled: true,
             concurrency_limit: None,
             weight: AccountWeight::DEFAULT,
+            model_access: super::AccountModelAccess::all(),
             credential_state: CredentialState::Unknown,
             quota: QuotaState::unknown(),
             last_error_reason: None,
@@ -728,6 +730,17 @@ impl ProviderAccount {
         self
     }
 
+    #[must_use]
+    pub fn with_model_access(mut self, model_access: super::AccountModelAccess) -> Self {
+        self.model_access = model_access;
+        self
+    }
+
+    #[must_use]
+    pub const fn model_access(&self) -> &super::AccountModelAccess {
+        &self.model_access
+    }
+
     #[must_use]
     pub fn with_outbound_proxy(mut self, proxy: Option<super::OutboundProxy>) -> Self {
         self.outbound_proxy = proxy;
@@ -1007,6 +1020,8 @@ impl ProviderRefreshQuery {
 #[derive(Clone, PartialEq)]
 pub struct NewProviderAccount {
     pub account: ProviderAccount,
+    /// 导入时显式提供的政策；省略时保留已有账号设置。
+    pub model_access: Option<super::AccountModelAccess>,
     pub credential: PlaintextCredential,
 }
 

backend/crates/gateway-core/src/account/model_access.rs

@@ -0,0 +1,105 @@
+//! 管理员配置的账号模型政策，与上游模型权限分开保存和判断。
+
+use std::collections::BTreeSet;
+
+use serde::{Deserialize, Deserializer, Serialize};
+use thiserror::Error;
+
+use crate::validation::validate_text;
+
+/// 一个账号最多配置的精确模型 ID 数量。
+pub const MAX_ACCOUNT_ACCESS_MODELS: usize = 256;
+
+/// 账号模型政策的匹配方式。
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "snake_case")]
+pub enum AccountModelAccessMode {
+    #[default]
+    All,
+    Allowlist,
+    Denylist,
+}
+
+/// 已校验的账号模型政策；私有字段保证所有边界使用相同的匹配规则。
+#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
+pub struct AccountModelAccess {
+    mode: AccountModelAccessMode,
+    models: BTreeSet<String>,
+}
+
+/// 不携带原始输入，避免在错误边界回显配置正文。
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Error)]
+#[error(
+    "modelAccess requires all with no models, or allowlist/denylist with 1–256 exact model IDs (at most 256 bytes each)"
+)]
+pub struct InvalidAccountModelAccess;
+
+impl Default for AccountModelAccess {
+    fn default() -> Self {
+        Self::all()
+    }
+}
+
+impl AccountModelAccess {
+    #[must_use]
+    pub const fn all() -> Self {
+        Self {
+            mode: AccountModelAccessMode::All,
+            models: BTreeSet::new(),
+        }
+    }
+
+    pub fn new(
+        mode: AccountModelAccessMode,
+        models: Vec<String>,
+    ) -> Result<Self, InvalidAccountModelAccess> {
+        if models.len() > MAX_ACCOUNT_ACCESS_MODELS
+            || (mode == AccountModelAccessMode::All) != models.is_empty()
+            || models.iter().any(|model| {
+                model.trim() != model
+                    || model.trim().is_empty()
+                    || model.contains('*')
+                    || validate_text(model, 256, true, None).is_err()
+            })
+        {
+            return Err(InvalidAccountModelAccess);
+        }
+        Ok(Self {
+            mode,
+            models: models.into_iter().collect(),
+        })
+    }
+
+    #[must_use]
+    pub const fn mode(&self) -> AccountModelAccessMode {
+        self.mode
+    }
+
+    #[must_use]
+    pub fn models(&self) -> &BTreeSet<String> {
+        &self.models
+    }
+
+    #[must_use]
+    pub fn allows(&self, upstream_model: &str) -> bool {
+        match self.mode {
+            AccountModelAccessMode::All => true,
+            AccountModelAccessMode::Allowlist => self.models.contains(upstream_model),
+            AccountModelAccessMode::Denylist => !self.models.contains(upstream_model),
+        }
+    }
+}
+
+#[derive(Deserialize)]
+#[serde(deny_unknown_fields)]
+struct ModelAccessDocument {
+    mode: AccountModelAccessMode,
+    models: Vec<String>,
+}
+
+impl<'de> Deserialize<'de> for AccountModelAccess {
+    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> Result<Self, D::Error> {
+        let document = ModelAccessDocument::deserialize(deserializer)?;
+        Self::new(document.mode, document.models).map_err(serde::de::Error::custom)
+    }
+}

backend/crates/gateway-core/src/account/scope.rs

@@ -47,6 +47,7 @@ impl fmt::Display for AccountGroupId {
 pub struct RuntimeAccount {
     provider_kind: ProviderKind,
     group_ids: Arc<BTreeSet<AccountGroupId>>,
+    model_access: super::AccountModelAccess,
 }
 
 impl RuntimeAccount {
@@ -55,9 +56,21 @@ impl RuntimeAccount {
         Self {
             provider_kind,
             group_ids: Arc::new(group_ids),
+            model_access: super::AccountModelAccess::all(),
         }
     }
 
+    #[must_use]
+    pub fn with_model_access(mut self, model_access: super::AccountModelAccess) -> Self {
+        self.model_access = model_access;
+        self
+    }
+
+    #[must_use]
+    pub const fn model_access(&self) -> &super::AccountModelAccess {
+        &self.model_access
+    }
+
     #[must_use]
     pub const fn provider_kind(&self) -> &ProviderKind {
         &self.provider_kind
@@ -266,6 +279,23 @@ impl FrozenAccountScope {
         }
     }
 
+    #[must_use]
+    pub fn allows_model(&self, account_id: &ProviderAccountId, upstream_model: &str) -> bool {
+        self.allows(account_id)
+            && self
+                .directory
+                .account(account_id)
+                .is_some_and(|account| account.model_access.allows(upstream_model))
+    }
+
+    /// 目录按整个授权账号池过滤，不能只看用于获取元数据的账号政策。
+    #[must_use]
+    pub fn allows_provider_model(&self, provider: &ProviderKind, upstream_model: &str) -> bool {
+        self.directory.accounts.iter().any(|(id, account)| {
+            account.provider_kind() == provider && self.allows_model(id, upstream_model)
+        })
+    }
+
     #[must_use]
     pub fn provider_kinds(&self) -> &BTreeSet<ProviderKind> {
         match &self.client_scope {

backend/crates/gateway-core/src/account/selection.rs

@@ -7,6 +7,7 @@ use std::sync::atomic::{AtomicU64, Ordering};
 use std::sync::{Mutex, RwLock};
 use std::time::{Duration, Instant, SystemTime};
 
+use crate::concurrency::ConcurrencyQueuePolicy;
 use crate::identity::ProviderKind;
 
 use super::{AccountStatus, ProviderAccount, ProviderAccountId};
@@ -49,6 +50,7 @@ pub struct AccountSelectionPolicy {
     strategy: RotationStrategy,
     max_concurrent_per_account: NonZeroU32,
     request_interval: Duration,
+    queue_policy: ConcurrencyQueuePolicy,
 }
 
 impl AccountSelectionPolicy {
@@ -62,9 +64,24 @@ impl AccountSelectionPolicy {
             strategy,
             max_concurrent_per_account,
             request_interval,
+            queue_policy: ConcurrencyQueuePolicy {
+                max_waiting: 0,
+                timeout: Duration::ZERO,
+            },
         }
     }
 
+    #[must_use]
+    pub const fn with_queue(mut self, policy: ConcurrencyQueuePolicy) -> Self {
+        self.queue_policy = policy;
+        self
+    }
+
+    #[must_use]
+    pub const fn queue_policy(self) -> ConcurrencyQueuePolicy {
+        self.queue_policy
+    }
+
     #[must_use]
     pub const fn strategy(self) -> RotationStrategy {
         self.strategy
@@ -590,6 +607,38 @@ impl AccountSelector {
         })
     }
 
+    /// 只有本地并发/调度间隔可等待；账号权限、失效、额度与上游冷却仍立即排除。
+    #[must_use]
+    pub fn wait_candidates(
+        &self,
+        candidates: &[AccountCandidate],
+        context: &AccountSelectionContext,
+    ) -> Vec<ProviderAccountId> {
+        let mut candidates = candidates
+            .iter()
+            .filter(|candidate| {
+                matches!(
+                    self.scheduling_blocker(candidate, context),
+                    None | Some(
+                        AccountSchedulingBlocker::ConcurrencyLimit
+                            | AccountSchedulingBlocker::RequestInterval
+                    )
+                )
+            })
+            .collect::<Vec<_>>();
+        candidates.sort_by_key(|candidate| {
+            (
+                context.preferred_account.as_ref() != Some(candidate.account.id()),
+                Reverse(candidate.account.weight()),
+                candidate.account.id().clone(),
+            )
+        });
+        candidates
+            .into_iter()
+            .map(|candidate| candidate.account.id().clone())
+            .collect()
+    }
+
     pub(crate) fn scheduling_blocker(
         &self,
         candidate: &AccountCandidate,

backend/crates/gateway-core/src/concurrency.rs

@@ -0,0 +1,285 @@
+//! 随请求存活的有界等待位置；执行容量仍由原子准入/租约端口裁决。
+
+use std::collections::{HashMap, VecDeque};
+use std::hash::Hash;
+use std::sync::{Arc, Mutex, OnceLock};
+use std::task::Poll;
+use std::time::{Duration, Instant, SystemTime};
+
+use futures::{FutureExt, future::poll_fn, pin_mut, select_biased, task::AtomicWaker};
+use futures_timer::Delay;
+
+const MAX_TOTAL_WAITING: usize = 1_024;
+const CAPACITY_RECHECK_INTERVAL: Duration = Duration::from_millis(100);
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub struct ConcurrencyQueuePolicy {
+    pub max_waiting: u32,
+    pub timeout: Duration,
+}
+
+/// 从首次入队开始计时，密钥、账号与后续重试共享同一个截止时刻。
+#[derive(Debug, Clone, Default)]
+pub struct ConcurrencyWaitBudget {
+    deadline: Arc<OnceLock<Instant>>,
+}
+
+impl ConcurrencyWaitBudget {
+    fn deadline(&self, timeout: Duration, request_deadline: SystemTime) -> Instant {
+        let now = Instant::now();
+        let remaining = request_deadline
+            .duration_since(SystemTime::now())
+            .unwrap_or_default();
+        let deadline = *self.deadline.get_or_init(|| now + timeout);
+        deadline.min(now + remaining)
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, thiserror::Error)]
+pub enum QueueRejection {
+    #[error("concurrency wait queue is full")]
+    Full,
+    #[error("concurrency wait deadline elapsed")]
+    Timeout,
+}
+
+struct QueueState<K> {
+    queues: HashMap<K, VecDeque<Arc<AtomicWaker>>>,
+    total: usize,
+}
+
+/// 只保存等待者与唤醒器，既不持有请求正文，也不复制运行中并发计数。
+pub struct ConcurrencyWaitQueue<K> {
+    state: Arc<Mutex<QueueState<K>>>,
+}
+
+impl<K> Default for ConcurrencyWaitQueue<K> {
+    fn default() -> Self {
+        Self {
+            state: Arc::new(Mutex::new(QueueState {
+                queues: HashMap::new(),
+                total: 0,
+            })),
+        }
+    }
+}
+
+impl<K: Clone + Eq + Hash> ConcurrencyWaitQueue<K> {
+    #[must_use]
+    pub fn has_waiters(&self, key: &K) -> bool {
+        self.state
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner)
+            .queues
+            .get(key)
+            .is_some_and(|queue| !queue.is_empty())
+    }
+
+    pub fn enqueue(
+        &self,
+        keys: &[K],
+        max_waiting: u32,
+        deadline: Instant,
+    ) -> Result<WaitTicket<K>, QueueRejection> {
+        if Instant::now() >= deadline {
+            return Err(QueueRejection::Timeout);
+        }
+        let mut state = self
+            .state
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        if state.total >= MAX_TOTAL_WAITING {
+            return Err(QueueRejection::Full);
+        }
+        let key = keys
+            .iter()
+            .filter_map(|key| {
+                let len = state.queues.get(key).map_or(0, VecDeque::len);
+                (len < max_waiting as usize).then_some((key, len))
+            })
+            .min_by_key(|(_, len)| *len)
+            .map(|(key, _)| key.clone())
+            .ok_or(QueueRejection::Full)?;
+        let waker = Arc::new(AtomicWaker::new());
+        state
+            .queues
+            .entry(key.clone())
+            .or_default()
+            .push_back(Arc::clone(&waker));
+        state.total += 1;
+        Ok(WaitTicket {
+            state: Arc::clone(&self.state),
+            key,
+            waker,
+            deadline,
+        })
+    }
+}
+
+/// Drop 同步回收等待位置，包括尚未轮到队首的取消请求。
+pub struct WaitTicket<K: Eq + Hash> {
+    state: Arc<Mutex<QueueState<K>>>,
+    key: K,
+    waker: Arc<AtomicWaker>,
+    deadline: Instant,
+}
+
+impl<K: Eq + Hash> WaitTicket<K> {
+    #[must_use]
+    pub const fn key(&self) -> &K {
+        &self.key
+    }
+
+    pub async fn turn(&self) -> Result<(), QueueRejection> {
+        let ready = poll_fn(|cx| {
+            self.waker.register(cx.waker());
+            let state = self
+                .state
+                .lock()
+                .unwrap_or_else(std::sync::PoisonError::into_inner);
+            if state
+                .queues
+                .get(&self.key)
+                .and_then(VecDeque::front)
+                .is_some_and(|head| Arc::ptr_eq(head, &self.waker))
+            {
+                Poll::Ready(())
+            } else {
+                Poll::Pending
+            }
+        })
+        .fuse();
+        let timeout = Delay::new(self.deadline.saturating_duration_since(Instant::now())).fuse();
+        pin_mut!(ready, timeout);
+        select_biased! {
+            _ = timeout => Err(QueueRejection::Timeout),
+            () = ready => if Instant::now() < self.deadline { Ok(()) } else { Err(QueueRejection::Timeout) },
+        }
+    }
+
+    pub async fn retry(&self) -> Result<(), QueueRejection> {
+        // 租约释放可能经过后台队列；有界重查同时覆盖释放通知之前及 TTL 到期的空位。
+        Delay::new(
+            CAPACITY_RECHECK_INTERVAL.min(self.deadline.saturating_duration_since(Instant::now())),
+        )
+        .await;
+        self.turn().await
+    }
+}
+
+impl<K: Eq + Hash> Drop for WaitTicket<K> {
+    fn drop(&mut self) {
+        let mut state = self
+            .state
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        let Some(queue) = state.queues.get_mut(&self.key) else {
+            return;
+        };
+        let Some(index) = queue.iter().position(|item| Arc::ptr_eq(item, &self.waker)) else {
+            return;
+        };
+        queue.remove(index);
+        let next = queue.front().cloned();
+        if queue.is_empty() {
+            state.queues.remove(&self.key);
+        }
+        state.total -= 1;
+        drop(state);
+        if index == 0
+            && let Some(next) = next
+        {
+            next.wake();
+        }
+    }
+}
+
+/// 单层等待位置与观测；时间预算由请求共享，重选账号不会重新开始计时。
+pub struct CapacityWait<'a, K: Eq + Hash> {
+    queue: &'a ConcurrencyWaitQueue<K>,
+    policy: ConcurrencyQueuePolicy,
+    request_deadline: SystemTime,
+    budget: &'a ConcurrencyWaitBudget,
+    started_at: Option<Instant>,
+    ticket: Option<WaitTicket<K>>,
+}
+
+impl<'a, K: Clone + Eq + Hash> CapacityWait<'a, K> {
+    #[must_use]
+    pub const fn new(
+        queue: &'a ConcurrencyWaitQueue<K>,
+        policy: ConcurrencyQueuePolicy,
+        request_deadline: SystemTime,
+        budget: &'a ConcurrencyWaitBudget,
+    ) -> Self {
+        Self {
+            queue,
+            policy,
+            request_deadline,
+            budget,
+            started_at: None,
+            ticket: None,
+        }
+    }
+
+    #[must_use]
+    pub fn can_try(&self, key: &K) -> bool {
+        self.ticket
+            .as_ref()
+            .is_some_and(|ticket| ticket.key() == key)
+            || !self.queue.has_waiters(key)
+    }
+
+    #[must_use]
+    pub fn elapsed(&self) -> Duration {
+        self.started_at
+            .map_or(Duration::ZERO, |start| start.elapsed())
+    }
+
+    pub async fn wait(&mut self, keys: &[K]) -> Result<(), QueueRejection> {
+        self.started_at.get_or_insert_with(Instant::now);
+        let deadline = self
+            .budget
+            .deadline(self.policy.timeout, self.request_deadline);
+        if self
+            .ticket
+            .as_ref()
+            .is_some_and(|ticket| !keys.contains(ticket.key()))
+        {
+            self.ticket = None;
+        }
+        if self.ticket.is_none() {
+            self.ticket = Some(
+                self.queue
+                    .enqueue(keys, self.policy.max_waiting, deadline)?,
+            );
+        }
+        if let Some(ticket) = &self.ticket {
+            ticket.retry().await?;
+        }
+        Ok(())
+    }
+}
+
+impl QueueRejection {
+    #[must_use]
+    pub const fn provider_kind(self) -> crate::error::ProviderErrorKind {
+        match self {
+            Self::Full => crate::error::ProviderErrorKind::ConcurrencyQueueFull,
+            Self::Timeout => crate::error::ProviderErrorKind::ConcurrencyQueueTimeout,
+        }
+    }
+    #[must_use]
+    pub fn gateway_error(self) -> crate::error::GatewayError {
+        match self {
+            Self::Full => crate::error::GatewayError::new(
+                crate::error::GatewayErrorKind::ConcurrencyQueueFull,
+                "concurrency wait queue is full",
+            ),
+            Self::Timeout => crate::error::GatewayError::new(
+                crate::error::GatewayErrorKind::ConcurrencyQueueTimeout,
+                "concurrency wait deadline elapsed",
+            ),
+        }
+    }
+}

backend/crates/gateway-core/src/engine/admission.rs

@@ -13,6 +13,8 @@ pub struct ClientAdmissionRequest {
     pub model_request_id: ModelRequestId,
     pub client_api_key_id: ClientApiKeyId,
     pub lease_ttl: Duration,
+    /// 有等待者时只允许队首取得槽位，RPM 仍由原子准入检查。
+    pub allow_concurrency_acquire: bool,
     pub limits: RateLimits,
 }
 
@@ -58,6 +60,9 @@ pub struct ClientAdmissionRestoreResult {
 pub struct ClientAdmissionError;
 
 pub trait ClientAdmissionPort: Send + Sync {
+    /// 请求 future 被取消时移交幂等释放，具体实现拥有异步清理执行器。
+    fn abandon(&self, client_api_key_id: &ClientApiKeyId, model_request_id: &ModelRequestId);
+
     fn admit(
         &self,
         request: ClientAdmissionRequest,

backend/crates/gateway-core/src/engine/coordinator.rs

@@ -1,5 +1,6 @@
 //! 唯一的账号重试、发送与下游 commit barrier owner。
 
+use crate::concurrency::ConcurrencyWaitBudget;
 use crate::diagnostics::TraceContext;
 use serde_json::json;
 
@@ -166,6 +167,7 @@ where
             engine: Arc::clone(&self.engine),
             request_id,
             client_api_key_ref,
+            concurrency_wait_budget: ConcurrencyWaitBudget::default(),
             observation: ResponseObservation::new(timing_started_at),
             budget_prior_attempts_usd: Decimal::ZERO,
             budget_attempt_already_counted: false,
@@ -270,6 +272,7 @@ pub struct ResponseExecutionSession<S: ?Sized> {
     engine: Arc<GatewayEngine<S>>,
     request_id: ModelRequestId,
     client_api_key_ref: crate::policy::ClientApiKeyId,
+    concurrency_wait_budget: ConcurrencyWaitBudget,
     observation: ResponseObservation,
     budget_prior_attempts_usd: Decimal,
     budget_attempt_already_counted: bool,
@@ -334,6 +337,12 @@ impl<S: ?Sized> ResponseExecutionSession<S>
 where
     S: ExecutionStore + 'static,
 {
+    /// 首次驱动前接入密钥准入使用的预算，所有账号尝试随后共享它。
+    pub(crate) fn with_concurrency_wait_budget(mut self, budget: ConcurrencyWaitBudget) -> Self {
+        self.concurrency_wait_budget = budget;
+        self
+    }
+
     /// 当前请求共享的诊断上下文。
     pub fn trace(&self) -> TraceContext {
         self.trace.clone()
@@ -779,6 +788,7 @@ where
         }));
         let context = AttemptContext::new(
             RequestAttemptContext::new(self.request_id.clone(), self.client_api_key_ref.clone())
+                .with_concurrency_wait_budget(self.concurrency_wait_budget.clone())
                 .with_timing_started_at(self.observation.timing_started_at)
                 .with_trace(self.trace.clone()),
             next_attempt,
@@ -848,6 +858,8 @@ where
                         ProviderErrorKind::AccountCapacityUnavailable
                             | ProviderErrorKind::NoEligibleAccount
                             | ProviderErrorKind::ProviderInfrastructureUnavailable
+                            | ProviderErrorKind::ConcurrencyQueueFull
+                            | ProviderErrorKind::ConcurrencyQueueTimeout
                     ) && error.send_state() == UpstreamSendState::NotSent
                         && matches!(
                             self.continuation_attempt,
@@ -861,6 +873,8 @@ where
                         error.kind(),
                         ProviderErrorKind::AccountCapacityUnavailable
                             | ProviderErrorKind::NoEligibleAccount
+                            | ProviderErrorKind::ConcurrencyQueueFull
+                            | ProviderErrorKind::ConcurrencyQueueTimeout
                     ) && let Some(last_failure) = self.last_retryable_failure.take()
                     {
                         let send_state = self.current_send_state();
@@ -882,6 +896,8 @@ where
                         ProviderErrorKind::AccountCapacityUnavailable
                             | ProviderErrorKind::NoEligibleAccount
                             | ProviderErrorKind::ProviderInfrastructureUnavailable
+                            | ProviderErrorKind::ConcurrencyQueueFull
+                            | ProviderErrorKind::ConcurrencyQueueTimeout
                     ) && error.send_state() == UpstreamSendState::NotSent)
                     {
                         self.record_provider_failure(candidate.provider().clone(), error.kind());

backend/crates/gateway-core/src/engine/execution.rs

@@ -11,6 +11,7 @@ use futures::{FutureExt, future::BoxFuture, pin_mut, select_biased};
 use futures_timer::Delay;
 use uuid::Uuid;
 
+use crate::concurrency::{CapacityWait, ConcurrencyWaitBudget, ConcurrencyWaitQueue};
 use crate::engine::admission::{
     ClientAdmissionDecision, ClientAdmissionPort, ClientAdmissionRejection, ClientAdmissionRequest,
 };
@@ -207,6 +208,16 @@ pub trait ExecutionService: Send + Sync {
     ) -> BoxFuture<'_, Result<StartedExecution, GatewayError>>;
 }
 
+/// 只验证 Client API Key 并返回稳定 Key ID，不产生请求使用事实。
+///
+/// 管理侧登录等非数据面场景必须使用该端口，避免把认证本身误记为一次 Key 使用。
+pub trait ClientKeyVerifier: Send + Sync {
+    fn verify_client_key(
+        &self,
+        plaintext: &str,
+    ) -> Result<ClientApiKeyId, ClientAuthenticationError>;
+}
+
 /// 成功认证后的 API Key 使用事实接收器。
 ///
 /// 认证仍是同步快照读取；实现必须自行异步、去重地持久化，不得阻塞客户端请求。
@@ -263,6 +274,7 @@ pub struct DefaultExecutionService {
     observations: Arc<dyn ExecutionStore>,
     providers: ProviderRegistry,
     admissions: Arc<dyn ClientAdmissionPort>,
+    admission_waiting: ConcurrencyWaitQueue<ClientApiKeyId>,
     circuits: Arc<dyn ProviderCircuitPort>,
     continuation: Arc<dyn NativeContinuationPort>,
     client_api_key_usage: Arc<dyn ClientApiKeyUsageSink>,
@@ -291,6 +303,7 @@ impl DefaultExecutionService {
             observations,
             providers,
             admissions,
+            admission_waiting: ConcurrencyWaitQueue::default(),
             circuits,
             continuation,
             client_api_key_usage,
@@ -304,6 +317,25 @@ impl DefaultExecutionService {
         self
     }
 
+    fn authenticate_without_usage(
+        &self,
+        plaintext: &str,
+    ) -> Result<AuthenticatedClient, ClientAuthenticationError> {
+        let snapshot = self
+            .snapshots
+            .acquire()
+            .map_err(|_| ClientAuthenticationError::SnapshotUnavailable)?;
+        let policy = snapshot
+            .client_policies()
+            .filter(|policy| {
+                constant_time_equal(plaintext, policy.plaintext_key().expose_for_auth())
+            })
+            .find(|policy| policy.authorize().is_ok())
+            .cloned()
+            .ok_or(ClientAuthenticationError::InvalidKey)?;
+        Ok(AuthenticatedClient { snapshot, policy })
+    }
+
     async fn start_inner(&self, request: StartExecution) -> Result<StartedExecution, GatewayError> {
         let StartExecution {
             client,
@@ -497,40 +529,12 @@ impl DefaultExecutionService {
             operation,
             metadata,
         } = request;
-        let admission_request = ClientAdmissionRequest {
-            model_request_id: request_id.clone(),
-            client_api_key_id: client.policy.key_id().clone(),
-            lease_ttl: MODEL_REQUEST_DEADLINE,
-            limits: client.policy.limits(),
-        };
         let admission_started_at = Instant::now();
-        match self
-            .admissions
-            .admit(admission_request)
-            .await
-            .map_err(|_| {
-                GatewayError::new(
-                    GatewayErrorKind::NoAvailableProvider,
-                    "request admission is temporarily unavailable",
-                )
-            })? {
-            ClientAdmissionDecision::Granted => {}
-            ClientAdmissionDecision::Rejected(
-                ClientAdmissionRejection::RateLimited
-                | ClientAdmissionRejection::ConcurrencyLimited,
-            ) => {
-                return Err(GatewayError::new(
-                    GatewayErrorKind::RateLimited,
-                    "request exceeds client API key limits",
-                ));
-            }
-        }
+        let concurrency_wait_budget = ConcurrencyWaitBudget::default();
+        let admission = self
+            .acquire_client_admission(&client, &request_id, deadline_at, &concurrency_wait_budget)
+            .await?;
         let admission_decision_ms = duration_ms(admission_started_at.elapsed());
-        let admission = AdmissionLease {
-            port: Arc::clone(&self.admissions),
-            client_api_key_id: client.policy.key_id().clone(),
-            model_request_id: request_id.clone(),
-        };
         if let Some(budget) = &self.budget
             && let Err(error) = budget.admit(client.policy.key_id().clone()).await
         {
@@ -583,7 +587,7 @@ impl DefaultExecutionService {
             )
             .await
         {
-            Ok(core) => core,
+            Ok(core) => core.with_concurrency_wait_budget(concurrency_wait_budget),
             Err(error) => {
                 if let Some(budget) = &self.budget {
                     settle_budget(
@@ -615,6 +619,85 @@ impl DefaultExecutionService {
         })
     }
 
+    async fn acquire_client_admission(
+        &self,
+        client: &AuthenticatedClient,
+        request_id: &ModelRequestId,
+        deadline_at: SystemTime,
+        budget: &ConcurrencyWaitBudget,
+    ) -> Result<AdmissionLease, GatewayError> {
+        let policy = client.snapshot.client_queue_policy();
+        let limits = client.policy.limits();
+        let key = client.policy.key_id();
+        let mut waiting = CapacityWait::new(&self.admission_waiting, policy, deadline_at, budget);
+        let mut admission = AdmissionLease {
+            port: Arc::clone(&self.admissions),
+            client_api_key_id: key.clone(),
+            model_request_id: request_id.clone(),
+            armed: false,
+        };
+        loop {
+            let remaining = deadline_at
+                .duration_since(SystemTime::now())
+                .unwrap_or_default();
+            if remaining.is_zero() {
+                return Err(GatewayError::new(
+                    GatewayErrorKind::Timeout,
+                    "request deadline elapsed",
+                ));
+            }
+            // 在发送原子准入之前接管取消清理，覆盖 Redis 已取得租约但返回尚未被观察的窗口。
+            admission.armed = true;
+            let acquire = self
+                .admissions
+                .admit(ClientAdmissionRequest {
+                    model_request_id: request_id.clone(),
+                    client_api_key_id: key.clone(),
+                    lease_ttl: remaining,
+                    allow_concurrency_acquire: limits.max_concurrency == 0 || waiting.can_try(key),
+                    limits,
+                })
+                .fuse();
+            let timeout = Delay::new(remaining).fuse();
+            pin_mut!(acquire, timeout);
+            let decision = select_biased! {
+                result = acquire => result.map_err(|_| GatewayError::new(GatewayErrorKind::NoAvailableProvider, "request admission is temporarily unavailable"))?,
+                _ = timeout => return Err(GatewayError::new(GatewayErrorKind::Timeout, "request deadline elapsed")),
+            };
+            match decision {
+                ClientAdmissionDecision::Granted => {
+                    if !waiting.elapsed().is_zero() {
+                        tracing::info!(
+                            request_id = request_id.as_str(),
+                            queue_layer = "client_key",
+                            queue_wait_ms = duration_ms(waiting.elapsed()),
+                            "排队请求已取得 Key 并发槽位"
+                        );
+                    }
+                    return Ok(admission);
+                }
+                ClientAdmissionDecision::Rejected(reason) => {
+                    admission.armed = false;
+                    if reason == ClientAdmissionRejection::RateLimited || policy.max_waiting == 0 {
+                        return Err(GatewayError::new(
+                            GatewayErrorKind::RateLimited,
+                            "request exceeds client API key limits",
+                        ));
+                    }
+                    if waiting.elapsed().is_zero()
+                        && let Some(budget) = &self.budget
+                    {
+                        budget.admit(key.clone()).await?;
+                    }
+                    waiting.wait(std::slice::from_ref(key)).await.map_err(|error| {
+                        tracing::info!(request_id = request_id.as_str(), queue_layer = "client_key", queue_wait_ms = duration_ms(waiting.elapsed()), reason = %error, "Key 排队请求被拒绝");
+                        error.gateway_error()
+                    })?;
+                }
+            }
+        }
+    }
+
     async fn route_context(
         &self,
         provider_kinds: &BTreeSet<ProviderKind>,
@@ -904,20 +987,10 @@ impl ExecutionService for DefaultExecutionService {
         &self,
         plaintext: &str,
     ) -> Result<AuthenticatedClient, ClientAuthenticationError> {
-        let snapshot = self
-            .snapshots
-            .acquire()
-            .map_err(|_| ClientAuthenticationError::SnapshotUnavailable)?;
-        let policy = snapshot
-            .client_policies()
-            .filter(|policy| {
-                constant_time_equal(plaintext, policy.plaintext_key().expose_for_auth())
-            })
-            .find(|policy| policy.authorize().is_ok())
-            .cloned()
-            .ok_or(ClientAuthenticationError::InvalidKey)?;
-        self.client_api_key_usage.record_used(policy.key_id());
-        Ok(AuthenticatedClient { snapshot, policy })
+        let client = self.authenticate_without_usage(plaintext)?;
+        self.client_api_key_usage
+            .record_used(client.policy().key_id());
+        Ok(client)
     }
 
     fn public_models(&self, client: &AuthenticatedClient) -> Vec<PublicModelId> {
@@ -943,6 +1016,12 @@ impl ExecutionService for DefaultExecutionService {
                     .await?
                 else {
                     for profile in client.snapshot.public_model_profiles_for_provider(kind) {
+                        if !scope.allows_provider_model(
+                            kind,
+                            &client.snapshot.mapped_model(profile.model().as_str()),
+                        ) {
+                            continue;
+                        }
                         if seen.insert(profile.model().clone()) {
                             result.push(PublicModelDescriptor::Adapted(profile));
                         }
@@ -956,6 +1035,9 @@ impl ExecutionService for DefaultExecutionService {
                     .collect::<std::collections::BTreeMap<_, _>>();
                 for entry in &models {
                     let target = client.snapshot.mapped_model(entry.model.as_str());
+                    if !scope.allows_provider_model(kind, &target) {
+                        continue;
+                    }
                     let Some(source) = by_id.get(target.as_str()) else {
                         continue;
                     };
@@ -977,6 +1059,9 @@ impl ExecutionService for DefaultExecutionService {
                 }
                 for model in client.snapshot.public_models_for_provider(kind) {
                     let target = client.snapshot.mapped_model(model.as_str());
+                    if !scope.allows_provider_model(kind, &target) {
+                        continue;
+                    }
                     if target == model.as_str() || seen.contains(&model) {
                         continue;
                     }
@@ -1020,6 +1105,16 @@ impl ExecutionService for DefaultExecutionService {
     }
 }
 
+impl ClientKeyVerifier for DefaultExecutionService {
+    fn verify_client_key(
+        &self,
+        plaintext: &str,
+    ) -> Result<ClientApiKeyId, ClientAuthenticationError> {
+        self.authenticate_without_usage(plaintext)
+            .map(|client| client.policy().key_id().clone())
+    }
+}
+
 impl AccountProbe for DefaultExecutionService {
     fn probe(
         &self,
@@ -1030,6 +1125,7 @@ impl AccountProbe for DefaultExecutionService {
 }
 
 struct AdmissionLease {
+    armed: bool,
     port: Arc<dyn ClientAdmissionPort>,
     client_api_key_id: ClientApiKeyId,
     model_request_id: ModelRequestId,
@@ -1042,7 +1138,7 @@ async fn settle_budget(port: &dyn ClientBudgetPort, charge: ClientBudgetCharge)
 }
 
 impl AdmissionLease {
-    async fn release(self) {
+    async fn release(mut self) {
         if let Err(error) = self
             .port
             .release(&self.client_api_key_id, &self.model_request_id)
@@ -1050,6 +1146,16 @@ impl AdmissionLease {
         {
             tracing::warn!(%error, "Client admission 释放失败，依赖租约 TTL 收敛");
         }
+        self.armed = false;
+    }
+}
+
+impl Drop for AdmissionLease {
+    fn drop(&mut self) {
+        if self.armed {
+            self.port
+                .abandon(&self.client_api_key_id, &self.model_request_id);
+        }
     }
 }
 

backend/crates/gateway-core/src/engine/mod.rs

@@ -318,6 +318,7 @@ pub struct RequestAttemptContext {
     client_api_key_ref: ClientApiKeyId,
     timing_started_at: Instant,
     trace: crate::diagnostics::TraceContext,
+    concurrency_wait_budget: crate::concurrency::ConcurrencyWaitBudget,
 }
 
 impl RequestAttemptContext {
@@ -328,6 +329,7 @@ impl RequestAttemptContext {
             client_api_key_ref,
             timing_started_at: Instant::now(),
             trace: crate::diagnostics::TraceContext::default(),
+            concurrency_wait_budget: crate::concurrency::ConcurrencyWaitBudget::default(),
         }
     }
 
@@ -337,6 +339,16 @@ impl RequestAttemptContext {
         self
     }
 
+    /// 传递本次请求已经消耗的排队预算，不能在新 attempt 中重置。
+    #[must_use]
+    pub fn with_concurrency_wait_budget(
+        mut self,
+        budget: crate::concurrency::ConcurrencyWaitBudget,
+    ) -> Self {
+        self.concurrency_wait_budget = budget;
+        self
+    }
+
     /// 覆盖本次请求的单调计时原点。
     #[must_use]
     pub fn with_timing_started_at(mut self, timing_started_at: Instant) -> Self {
@@ -454,6 +466,12 @@ impl AttemptContext {
         self.deadline
     }
 
+    /// 返回密钥准入、账号选择与重试共用的排队预算。
+    #[must_use]
+    pub const fn concurrency_wait_budget(&self) -> &crate::concurrency::ConcurrencyWaitBudget {
+        &self.request.concurrency_wait_budget
+    }
+
     #[must_use]
     pub const fn account_selection_policy(&self) -> AccountSelectionPolicy {
         self.account_selection_policy

backend/crates/gateway-core/src/error.rs

@@ -34,6 +34,8 @@ pub enum ProviderErrorKind {
     QuotaExhausted,
     /// 仍有符合条件的账号，但它们暂时没有可调度容量。
     AccountCapacityUnavailable,
+    ConcurrencyQueueFull,
+    ConcurrencyQueueTimeout,
     /// Provider 已确认当前请求无法选出可用账号。
     NoEligibleAccount,
     /// Provider 的账号存储、租约协调或本地凭据数据不可用。
@@ -67,6 +69,8 @@ impl ProviderErrorKind {
             Self::RateLimited => "rate_limited",
             Self::QuotaExhausted => "quota_exhausted",
             Self::AccountCapacityUnavailable => "account_capacity_unavailable",
+            Self::ConcurrencyQueueFull => "concurrency_queue_full",
+            Self::ConcurrencyQueueTimeout => "concurrency_queue_timeout",
             Self::NoEligibleAccount => "no_eligible_account",
             Self::ProviderInfrastructureUnavailable => "provider_infrastructure_unavailable",
             Self::Timeout => "timeout",
@@ -960,6 +964,8 @@ pub enum GatewayErrorKind {
     NoAvailableProvider,
     /// 符合条件的上游账号暂时没有调度容量。
     AccountCapacityUnavailable,
+    ConcurrencyQueueFull,
+    ConcurrencyQueueTimeout,
     /// Provider 的本地账号基础设施不可用。
     ProviderInfrastructureUnavailable,
     /// 上游限流。
@@ -986,6 +992,8 @@ impl GatewayErrorKind {
             Self::ModelNotFound => "model_not_found",
             Self::NoAvailableProvider => "no_available_provider",
             Self::AccountCapacityUnavailable => "account_capacity_unavailable",
+            Self::ConcurrencyQueueFull => "concurrency_queue_full",
+            Self::ConcurrencyQueueTimeout => "concurrency_queue_timeout",
             Self::ProviderInfrastructureUnavailable => "provider_infrastructure_unavailable",
             Self::RateLimited => "rate_limited",
             Self::UpstreamUnavailable => "upstream_unavailable",
@@ -1072,6 +1080,14 @@ impl GatewayError {
                 GatewayErrorKind::RateLimited,
                 "upstream capacity is temporarily unavailable",
             ),
+            ProviderErrorKind::ConcurrencyQueueFull => Self::new(
+                GatewayErrorKind::ConcurrencyQueueFull,
+                "concurrency wait queue is full",
+            ),
+            ProviderErrorKind::ConcurrencyQueueTimeout => Self::new(
+                GatewayErrorKind::ConcurrencyQueueTimeout,
+                "concurrency wait deadline elapsed",
+            ),
             ProviderErrorKind::AccountCapacityUnavailable => Self::new(
                 GatewayErrorKind::AccountCapacityUnavailable,
                 "all eligible upstream accounts are temporarily busy",

backend/crates/gateway-core/src/lib.rs

@@ -4,6 +4,7 @@
 //! 具体客户端协议和具体 Provider 都通过外层 adapter 接入。
 
 pub mod account;
+pub mod concurrency;
 pub mod diagnostics;
 pub mod engine;
 pub mod error;
@@ -30,7 +31,8 @@ use engine::admission::{
 };
 use engine::continuation::NativeContinuationPort;
 use engine::execution::{
-    ClientApiKeyUsageSink, DefaultExecutionService, ExecutionService, ProviderCircuitPort,
+    ClientApiKeyUsageSink, ClientKeyVerifier, DefaultExecutionService, ExecutionService,
+    ProviderCircuitPort,
 };
 use engine::probe::AccountProbe;
 use engine::provider::ProviderRegistry;
@@ -93,6 +95,7 @@ impl CoreStorePorts {
 
 pub struct CoreBundle {
     execution: Arc<dyn ExecutionService>,
+    client_key_verifier: Arc<dyn ClientKeyVerifier>,
     snapshot_control: Arc<dyn SnapshotControl>,
     account_probe: Arc<dyn AccountProbe>,
     health_probes: Vec<Arc<dyn HealthProbe>>,
@@ -105,6 +108,11 @@ impl CoreBundle {
         Arc::clone(&self.execution)
     }
 
+    #[must_use]
+    pub fn client_key_verifier(&self) -> Arc<dyn ClientKeyVerifier> {
+        Arc::clone(&self.client_key_verifier)
+    }
+
     #[must_use]
     pub fn snapshot_control(&self) -> Arc<dyn SnapshotControl> {
         Arc::clone(&self.snapshot_control)
@@ -167,11 +175,13 @@ pub async fn initialize(
     }
     let service = Arc::new(service);
     let execution: Arc<dyn ExecutionService> = service.clone();
+    let client_key_verifier: Arc<dyn ClientKeyVerifier> = service.clone();
     let account_probe: Arc<dyn AccountProbe> = service;
     let snapshot_control: Arc<dyn SnapshotControl> = publisher;
     let health_probes: Vec<Arc<dyn HealthProbe>> = vec![Arc::new(snapshots)];
     Ok(CoreBundle {
         execution,
+        client_key_verifier,
         snapshot_control,
         account_probe,
         health_probes,

backend/crates/gateway-core/src/routing/snapshot.rs

@@ -8,6 +8,7 @@ use std::time::Duration;
 use futures::future::BoxFuture;
 
 use crate::account::{AccountSelectionPolicy, ProviderAccountId, RotationStrategy};
+use crate::concurrency::ConcurrencyQueuePolicy;
 use crate::operation::Operation;
 use crate::policy::{
     ClientApiKeyId, ClientPolicy, CodexClientMinVersions, CodexClientVersion,
@@ -28,6 +29,9 @@ const MAXIMUM_CATALOG_STABILITY_ATTEMPTS: usize = 4;
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct SnapshotSettingsFacts {
     max_concurrent_per_account: u32,
+    max_waiting_per_key: u32,
+    max_waiting_per_account: u32,
+    concurrency_wait_timeout_seconds: u32,
     request_interval_ms: u64,
     rotation_strategy: String,
     model_mappings: BTreeMap<String, String>,
@@ -36,6 +40,19 @@ pub struct SnapshotSettingsFacts {
 }
 
 impl SnapshotSettingsFacts {
+    #[must_use]
+    pub const fn with_concurrency_queues(
+        mut self,
+        max_waiting_per_key: u32,
+        max_waiting_per_account: u32,
+        timeout_seconds: u32,
+    ) -> Self {
+        self.max_waiting_per_key = max_waiting_per_key;
+        self.max_waiting_per_account = max_waiting_per_account;
+        self.concurrency_wait_timeout_seconds = timeout_seconds;
+        self
+    }
+
     #[must_use]
     pub fn new(
         max_concurrent_per_account: u32,
@@ -47,6 +64,9 @@ impl SnapshotSettingsFacts {
     ) -> Self {
         Self {
             max_concurrent_per_account,
+            max_waiting_per_key: 0,
+            max_waiting_per_account: 0,
+            concurrency_wait_timeout_seconds: 30,
             request_interval_ms,
             rotation_strategy: rotation_strategy.into(),
             model_mappings,
@@ -102,6 +122,7 @@ impl SnapshotAccountGroupFacts {
 pub struct SnapshotProviderAccountFacts {
     account_id: ProviderAccountId,
     provider_kind: String,
+    model_access: crate::account::AccountModelAccess,
 }
 
 impl SnapshotProviderAccountFacts {
@@ -110,10 +131,19 @@ impl SnapshotProviderAccountFacts {
         Self {
             account_id,
             provider_kind: provider_kind.into(),
+            model_access: crate::account::AccountModelAccess::all(),
         }
     }
 }
 
+impl SnapshotProviderAccountFacts {
+    #[must_use]
+    pub fn with_model_access(mut self, model_access: crate::account::AccountModelAccess) -> Self {
+        self.model_access = model_access;
+        self
+    }
+}
+
 /// Store 读取到的一条分组成员关系。
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct SnapshotAccountGroupMemberFacts {
@@ -307,7 +337,8 @@ async fn compile_runtime_snapshot(
             || accounts
                 .insert(
                     account.account_id.clone(),
-                    RuntimeAccount::new(provider_kind, BTreeSet::new()),
+                    RuntimeAccount::new(provider_kind, BTreeSet::new())
+                        .with_model_access(account.model_access),
                 )
                 .is_some()
         {
@@ -332,10 +363,23 @@ async fn compile_runtime_snapshot(
         let account = accounts
             .get_mut(&account_id)
             .ok_or(RuntimeSnapshotCompileError::InvalidData)?;
-        *account = RuntimeAccount::new(account.provider_kind().clone(), group_ids);
+        *account = RuntimeAccount::new(account.provider_kind().clone(), group_ids)
+            .with_model_access(account.model_access().clone());
     }
     let account_directory = Arc::new(RuntimeAccountDirectory::new(accounts));
 
+    if facts.settings.max_waiting_per_key > 1_000
+        || facts.settings.max_waiting_per_account > 1_000
+        || !(1..=120).contains(&facts.settings.concurrency_wait_timeout_seconds)
+    {
+        return Err(RuntimeSnapshotCompileError::InvalidData);
+    }
+    let queue_timeout =
+        Duration::from_secs(u64::from(facts.settings.concurrency_wait_timeout_seconds));
+    let client_queue_policy = ConcurrencyQueuePolicy {
+        max_waiting: facts.settings.max_waiting_per_key,
+        timeout: queue_timeout,
+    };
     let model_mappings = facts.settings.model_mappings;
     let min_client_versions = CodexClientMinVersions::new(
         facts
@@ -360,7 +404,11 @@ async fn compile_runtime_snapshot(
         NonZeroU32::new(facts.settings.max_concurrent_per_account)
             .ok_or(RuntimeSnapshotCompileError::InvalidData)?,
         Duration::from_millis(facts.settings.request_interval_ms),
-    );
+    )
+    .with_queue(ConcurrencyQueuePolicy {
+        max_waiting: facts.settings.max_waiting_per_account,
+        timeout: queue_timeout,
+    });
     let mut client_policies = Vec::with_capacity(facts.client_policies.len());
     for policy in facts.client_policies {
         let account_scope = if policy.group_ids.is_empty() {
@@ -414,6 +462,7 @@ async fn compile_runtime_snapshot(
     .map_err(|_| RuntimeSnapshotCompileError::InvalidData)
     .map(|snapshot| {
         snapshot
+            .with_client_queue_policy(client_queue_policy)
             .with_model_mappings(model_mappings)
             .with_account_directory(account_directory)
             .with_known_provider_catalogs(known_provider_catalogs)
@@ -425,6 +474,7 @@ async fn compile_runtime_snapshot(
 #[derive(Debug, Clone)]
 pub struct RuntimeSnapshot {
     revision: ConfigRevision,
+    client_queue_policy: ConcurrencyQueuePolicy,
     account_selection_policy: AccountSelectionPolicy,
     providers: Arc<BTreeSet<ProviderKind>>,
     provider_models: Arc<BTreeMap<ProviderKind, BTreeMap<UpstreamModelId, ModelCapabilities>>>,
@@ -439,6 +489,17 @@ pub struct RuntimeSnapshot {
 }
 
 impl RuntimeSnapshot {
+    #[must_use]
+    pub const fn client_queue_policy(&self) -> ConcurrencyQueuePolicy {
+        self.client_queue_policy
+    }
+
+    #[must_use]
+    pub const fn with_client_queue_policy(mut self, policy: ConcurrencyQueuePolicy) -> Self {
+        self.client_queue_policy = policy;
+        self
+    }
+
     /// 校验 Provider、实时模型目录和 Client API Key，并构建快照。
     pub fn new(
         revision: ConfigRevision,
@@ -509,6 +570,7 @@ impl RuntimeSnapshot {
         Ok(Self {
             revision,
             account_selection_policy,
+            client_queue_policy: ConcurrencyQueuePolicy::default(),
             providers: Arc::new(provider_set),
             provider_models: Arc::new(model_map),
             provider_model_presentations: Arc::new(presentation_map),
@@ -641,7 +703,13 @@ impl RuntimeSnapshot {
         scope
             .provider_kinds()
             .iter()
-            .flat_map(|provider| self.public_models_for_provider(provider))
+            .flat_map(|provider| {
+                self.public_models_for_provider(provider)
+                    .into_iter()
+                    .filter(|model| {
+                        scope.allows_provider_model(provider, &self.mapped_model(model.as_str()))
+                    })
+            })
             .collect::<BTreeSet<_>>()
             .into_iter()
             .collect()
@@ -656,6 +724,11 @@ impl RuntimeSnapshot {
         let mut profiles = BTreeMap::new();
         for provider in scope.provider_kinds() {
             for profile in self.public_model_profiles_for_provider(provider) {
+                if !scope
+                    .allows_provider_model(provider, &self.mapped_model(profile.model().as_str()))
+                {
+                    continue;
+                }
                 profiles
                     .entry(profile.model().clone())
                     .or_insert_with(|| profile.presentation().clone());
@@ -690,10 +763,10 @@ impl RuntimeSnapshot {
         public_model: &PublicModelId,
         scope: &FrozenAccountScope,
     ) -> bool {
-        scope
-            .provider_kinds()
-            .iter()
-            .any(|provider| self.contains_public_model_for_provider(public_model, provider))
+        scope.provider_kinds().iter().any(|provider| {
+            self.contains_public_model_for_provider(public_model, provider)
+                && scope.allows_provider_model(provider, &self.mapped_model(public_model.as_str()))
+        })
     }
 
     #[must_use]

backend/crates/gateway-core/tests/account/mod.rs

@@ -1,3 +1,4 @@
+mod model_access;
 mod proxy;
 mod selection;
 mod selection_diagnostics;

backend/crates/gateway-core/tests/account/model_access.rs

@@ -0,0 +1,36 @@
+use gateway_core::account::AccountModelAccess;
+use serde_json::json;
+
+#[test]
+fn account_model_access_document_rejects_invalid_or_unbounded_settings() {
+    for invalid in [
+        json!({"mode":"invalid","models":[]}),
+        json!({"mode":"all","models":["test-model"]}),
+        json!({"mode":"allowlist","models":[]}),
+        json!({"mode":"denylist","models":[]}),
+        json!({"mode":"allowlist","models":["gpt-*"]}),
+        json!({"mode":"allowlist","models":[" test-model"]}),
+        json!({"mode":"allowlist","models":["\n"]}),
+        json!({"mode":"allowlist","models":["__internal"]}),
+        json!({"mode":"allowlist","models":["x".repeat(257)]}),
+        json!({"mode":"allowlist","models":vec!["test-model";257]}),
+        json!({"mode":"all","models":[],"unknown":true}),
+        json!({"mode":"all"}),
+    ] {
+        assert!(serde_json::from_value::<AccountModelAccess>(invalid).is_err());
+    }
+}
+
+#[test]
+fn account_model_access_document_deduplicates_without_changing_exact_ids() {
+    let policy: AccountModelAccess = serde_json::from_value(json!({
+        "mode":"allowlist","models":["test-luna","test-luna","Test-Luna"]
+    }))
+    .expect("policy");
+    assert_eq!(
+        serde_json::to_value(policy).expect("serialize"),
+        json!({
+            "mode":"allowlist","models":["Test-Luna","test-luna"]
+        })
+    );
+}

backend/crates/gateway-core/tests/concurrency.rs

@@ -0,0 +1,161 @@
+use std::time::{Duration, Instant, SystemTime};
+
+use futures::{FutureExt, executor::block_on};
+use gateway_core::concurrency::{
+    CapacityWait, ConcurrencyQueuePolicy, ConcurrencyWaitBudget, ConcurrencyWaitQueue,
+    QueueRejection,
+};
+
+#[test]
+fn waiting_is_bounded_per_owner_and_cancelled_head_wakes_the_next_request() {
+    block_on(async {
+        let queue = ConcurrencyWaitQueue::default();
+        let deadline = Instant::now() + Duration::from_secs(2);
+        let first = queue.enqueue(&["key_a"], 2, deadline).unwrap();
+        let second = queue.enqueue(&["key_a"], 2, deadline).unwrap();
+        assert!(matches!(
+            queue.enqueue(&["key_a"], 2, deadline),
+            Err(QueueRejection::Full)
+        ));
+        let independent = queue.enqueue(&["key_b"], 2, deadline).unwrap();
+        assert!(second.turn().now_or_never().is_none());
+        independent.turn().await.unwrap();
+        first.turn().await.unwrap();
+        drop(first);
+        second.turn().await.unwrap();
+        drop(second);
+        assert!(!queue.has_waiters(&"key_a"));
+    });
+}
+
+#[test]
+fn cancelling_a_middle_waiter_preserves_fifo_and_reclaims_capacity() {
+    block_on(async {
+        let queue = ConcurrencyWaitQueue::default();
+        let deadline = Instant::now() + Duration::from_secs(2);
+        let head = queue.enqueue(&["account"], 3, deadline).unwrap();
+        let cancelled = queue.enqueue(&["account"], 3, deadline).unwrap();
+        let next = queue.enqueue(&["account"], 3, deadline).unwrap();
+        drop(cancelled);
+        let tail = queue.enqueue(&["account"], 3, deadline).unwrap();
+        drop(head);
+        next.turn().await.unwrap();
+        assert!(tail.turn().now_or_never().is_none());
+        drop(next);
+        tail.turn().await.unwrap();
+    });
+}
+
+#[test]
+fn wait_deadline_applies_before_and_after_becoming_head() {
+    block_on(async {
+        let queue = ConcurrencyWaitQueue::default();
+        let deadline = Instant::now() + Duration::from_millis(20);
+        let head = queue.enqueue(&["account"], 2, deadline).unwrap();
+        let tail = queue.enqueue(&["account"], 2, deadline).unwrap();
+        assert_eq!(tail.turn().await, Err(QueueRejection::Timeout));
+        assert_eq!(head.turn().await, Err(QueueRejection::Timeout));
+        drop((head, tail));
+        assert!(!queue.has_waiters(&"account"));
+    });
+}
+
+#[test]
+fn a_full_account_queue_does_not_prevent_waiting_on_another_eligible_account() {
+    let queue = ConcurrencyWaitQueue::default();
+    let deadline = Instant::now() + Duration::from_secs(2);
+    let _first = queue.enqueue(&["a"], 1, deadline).unwrap();
+    let second = queue.enqueue(&["a", "b"], 1, deadline).unwrap();
+    assert_eq!(*second.key(), "b");
+    assert!(matches!(
+        queue.enqueue(&["a", "b"], 1, deadline),
+        Err(QueueRejection::Full)
+    ));
+}
+
+#[test]
+fn request_deadline_bounds_waiting_and_dropped_future_releases_its_ticket() {
+    block_on(async {
+        let queue = ConcurrencyWaitQueue::default();
+        let policy = ConcurrencyQueuePolicy {
+            max_waiting: 1,
+            timeout: Duration::from_secs(30),
+        };
+        let budget = ConcurrencyWaitBudget::default();
+        let mut waiting = CapacityWait::new(
+            &queue,
+            policy,
+            SystemTime::now() + Duration::from_millis(20),
+            &budget,
+        );
+        assert_eq!(waiting.wait(&["a"]).await, Err(QueueRejection::Timeout));
+        drop(waiting);
+        assert!(!queue.has_waiters(&"a"));
+
+        let budget = ConcurrencyWaitBudget::default();
+        let mut waiting = CapacityWait::new(
+            &queue,
+            policy,
+            SystemTime::now() + Duration::from_secs(2),
+            &budget,
+        );
+        assert!(waiting.wait(&["a"]).now_or_never().is_none());
+        assert!(queue.has_waiters(&"a"));
+        drop(waiting);
+        assert!(!queue.has_waiters(&"a"));
+    });
+}
+
+#[test]
+fn new_layers_and_retries_cannot_restart_a_requests_wait_budget() {
+    block_on(async {
+        let keys = ConcurrencyWaitQueue::default();
+        let accounts = ConcurrencyWaitQueue::default();
+        let policy = ConcurrencyQueuePolicy {
+            max_waiting: 1,
+            timeout: Duration::from_millis(300),
+        };
+        let request_deadline = SystemTime::now() + Duration::from_secs(5);
+        let budget = ConcurrencyWaitBudget::default();
+        let mut key_wait = CapacityWait::new(&keys, policy, request_deadline, &budget);
+        key_wait.wait(&["key"]).await.unwrap();
+        drop(key_wait);
+
+        // 第一层已取得重试机会，后续准备跨过总时限后不能在账号层重新排队。
+        futures_timer::Delay::new(policy.timeout).await;
+        let next_attempt_budget = budget.clone();
+        let mut account_wait =
+            CapacityWait::new(&accounts, policy, request_deadline, &next_attempt_budget);
+        assert_eq!(
+            account_wait.wait(&["account"]).now_or_never(),
+            Some(Err(QueueRejection::Timeout))
+        );
+        assert!(!accounts.has_waiters(&"account"));
+
+        // 另一个请求仍有完整预算，不能受上一请求超时影响。
+        let independent_budget = ConcurrencyWaitBudget::default();
+        let mut independent =
+            CapacityWait::new(&accounts, policy, request_deadline, &independent_budget);
+        independent.wait(&["account"]).await.unwrap();
+    });
+}
+
+#[test]
+fn creating_a_waiter_does_not_start_the_budget_before_queueing() {
+    block_on(async {
+        let queue = ConcurrencyWaitQueue::default();
+        let policy = ConcurrencyQueuePolicy {
+            max_waiting: 1,
+            timeout: Duration::from_millis(300),
+        };
+        let budget = ConcurrencyWaitBudget::default();
+        let mut waiting = CapacityWait::new(
+            &queue,
+            policy,
+            SystemTime::now() + Duration::from_secs(5),
+            &budget,
+        );
+        futures_timer::Delay::new(policy.timeout).await;
+        waiting.wait(&["account"]).await.unwrap();
+    });
+}

backend/crates/gateway-core/tests/engine/admission.rs

@@ -183,6 +183,14 @@ struct RecordingAdmissions {
 }
 
 impl ClientAdmissionPort for RecordingAdmissions {
+    fn abandon(
+        &self,
+        key: &gateway_core::policy::ClientApiKeyId,
+        request: &gateway_core::engine::ModelRequestId,
+    ) {
+        let _ = futures::FutureExt::now_or_never(self.release(key, request));
+    }
+
     fn admit(
         &self,
         _: ClientAdmissionRequest,

backend/crates/gateway-core/tests/engine/coordinator.rs

@@ -3598,69 +3598,75 @@ fn transient_backoff_can_be_cancelled_before_the_next_attempt() {
 
 #[test]
 fn final_capacity_exhaustion_returns_the_last_retryable_upstream_failure() {
-    let operation = generate_operation();
-    let route_plan = plan(&operation);
-    let raw_body = Bytes::from_static(b"{\"error\":\"last upstream body\"}\x00");
-    let upstream_error =
-        ProviderError::new(ProviderErrorKind::Transport, UpstreamSendState::NotSent)
-            .with_pre_delivery_retry()
-            .with_client_visible_upstream_error(ClientVisibleUpstreamError::new(
-                "message too big",
-                Some("1009".to_owned()),
-                Some("websocket_close_error".to_owned()),
-            ))
-            .with_client_visible_upstream_response(ClientVisibleUpstreamResponse::new(
-                502,
-                Some(b"application/json".to_vec()),
-                raw_body.clone(),
-            ));
-    let (coordinator, store, provider) = coordinator(vec![
-        Script::Stream {
-            account_id: "acct_first",
-            items: vec![Err(upstream_error)],
-        },
-        Script::Error(ProviderError::new(
-            ProviderErrorKind::AccountCapacityUnavailable,
-            UpstreamSendState::NotSent,
-        )),
-    ]);
+    for capacity_error in [
+        ProviderErrorKind::AccountCapacityUnavailable,
+        ProviderErrorKind::ConcurrencyQueueFull,
+        ProviderErrorKind::ConcurrencyQueueTimeout,
+    ] {
+        let operation = generate_operation();
+        let route_plan = plan(&operation);
+        let raw_body = Bytes::from_static(b"{\"error\":\"last upstream body\"}\x00");
+        let upstream_error =
+            ProviderError::new(ProviderErrorKind::Transport, UpstreamSendState::NotSent)
+                .with_pre_delivery_retry()
+                .with_client_visible_upstream_error(ClientVisibleUpstreamError::new(
+                    "message too big",
+                    Some("1009".to_owned()),
+                    Some("websocket_close_error".to_owned()),
+                ))
+                .with_client_visible_upstream_response(ClientVisibleUpstreamResponse::new(
+                    502,
+                    Some(b"application/json".to_vec()),
+                    raw_body.clone(),
+                ));
+        let (coordinator, store, provider) = coordinator(vec![
+            Script::Stream {
+                account_id: "acct_first",
+                items: vec![Err(upstream_error)],
+            },
+            Script::Error(ProviderError::new(
+                capacity_error,
+                UpstreamSendState::NotSent,
+            )),
+        ]);
 
-    let mut session = block_on(coordinator.start(
-        model_request(&operation, SystemTime::now() + Duration::from_secs(30)),
-        operation,
-        route_plan,
-        None,
-        None,
-        CancellationToken::new(),
-    ))
-    .expect("start execution");
-    let error = block_on(session.collect_uncommitted())
-        .expect_err("empty selection must return the retained upstream failure");
-    let EngineError::Provider(error) = error else {
-        panic!("expected retained provider error");
-    };
+        let mut session = block_on(coordinator.start(
+            model_request(&operation, SystemTime::now() + Duration::from_secs(30)),
+            operation,
+            route_plan,
+            None,
+            None,
+            CancellationToken::new(),
+        ))
+        .expect("start execution");
+        let error = block_on(session.collect_uncommitted())
+            .expect_err("empty selection must return the retained upstream failure");
+        let EngineError::Provider(error) = error else {
+            panic!("expected retained provider error");
+        };
 
-    assert_eq!(error.kind(), ProviderErrorKind::Transport);
-    let detail = error
-        .client_visible_upstream_error()
-        .expect("structured upstream detail");
-    assert_eq!(detail.message(), "message too big");
-    assert_eq!(detail.code(), Some("1009"));
-    assert_eq!(detail.error_type(), Some("websocket_close_error"));
-    let response = error
-        .client_visible_upstream_response()
-        .expect("original upstream response");
-    assert_eq!(response.status(), 502);
-    assert_eq!(response.body(), &raw_body);
+        assert_eq!(error.kind(), ProviderErrorKind::Transport);
+        let detail = error
+            .client_visible_upstream_error()
+            .expect("structured upstream detail");
+        assert_eq!(detail.message(), "message too big");
+        assert_eq!(detail.code(), Some("1009"));
+        assert_eq!(detail.error_type(), Some("websocket_close_error"));
+        let response = error
+            .client_visible_upstream_response()
+            .expect("original upstream response");
+        assert_eq!(response.status(), 502);
+        assert_eq!(response.body(), &raw_body);
 
-    assert_eq!(provider.contexts.lock().expect("contexts lock").len(), 2);
-    let state = store.state.lock().expect("store lock");
-    assert_eq!(state.intermediate_failures, 1);
-    assert_eq!(state.finalizations[0].outcome, ExecutionOutcome::Failed);
-    assert_eq!(
-        state.finalizations[0].send_state,
-        UpstreamSendState::NotSent
-    );
+        assert_eq!(provider.contexts.lock().expect("contexts lock").len(), 2);
+        let state = store.state.lock().expect("store lock");
+        assert_eq!(state.intermediate_failures, 1);
+        assert_eq!(state.finalizations[0].outcome, ExecutionOutcome::Failed);
+        assert_eq!(
+            state.finalizations[0].send_state,
+            UpstreamSendState::NotSent
+        );
+    }
 }
 
 #[test]

backend/crates/gateway-core/tests/engine/execution.rs

@@ -19,6 +19,14 @@ struct Admissions {
 }
 
 impl ClientAdmissionPort for Admissions {
+    fn abandon(
+        &self,
+        key: &gateway_core::policy::ClientApiKeyId,
+        request: &gateway_core::engine::ModelRequestId,
+    ) {
+        let _ = futures::FutureExt::now_or_never(self.release(key, request));
+    }
+
     fn admit(
         &self,
         request: ClientAdmissionRequest,
@@ -814,9 +822,10 @@ use gateway_core::engine::continuation::{
     NativeContinuationPin, NativeContinuationPort, NativeContinuationStoreError, PreviousResponseId,
 };
 use gateway_core::engine::execution::{
-    ClientApiKeyUsageSink, ClientTransport, DefaultExecutionService, ExecutionRequestMetadata,
-    ExecutionService, ExecutionSession, ProviderCircuitDecision, ProviderCircuitError,
-    ProviderCircuitPort, StartExecution, StartProviderExecution, provider_failure_affects_circuit,
+    ClientApiKeyUsageSink, ClientKeyVerifier, ClientTransport, DefaultExecutionService,
+    ExecutionRequestMetadata, ExecutionService, ExecutionSession, ProviderCircuitDecision,
+    ProviderCircuitError, ProviderCircuitPort, StartExecution, StartProviderExecution,
+    provider_failure_affects_circuit,
 };
 use gateway_core::engine::probe::{AccountProbe, AccountProbeErrorSource, AccountProbeRequest};
 use gateway_core::engine::provider::{
@@ -1233,6 +1242,27 @@ fn successful_authentication_should_record_client_key_usage() {
     assert_eq!(usage.recorded(), vec!["key_usage_test".to_owned()]);
 }
 
+#[test]
+fn client_key_verification_should_not_record_client_key_usage() {
+    let usage = Arc::new(RecordingClientApiKeyUsage::default());
+    let service = DefaultExecutionService::new(
+        RuntimeSnapshotHandle::new(client_snapshot()),
+        Arc::new(TrackingExecutionStore::default()),
+        ProviderRegistry::default(),
+        Arc::new(UnusedAdmissions),
+        Arc::new(UnusedCircuits),
+        Arc::new(UnusedContinuation),
+        usage.clone(),
+    );
+
+    let key_id = service
+        .verify_client_key("sk_usage_test")
+        .expect("client key verification");
+
+    assert_eq!(key_id.as_str(), "key_usage_test");
+    assert!(usage.recorded().is_empty());
+}
+
 #[test]
 fn provider_endpoint_should_persist_its_real_v1_endpoint() {
     assert_provider_endpoint_observation(None);
@@ -1647,6 +1677,14 @@ impl ExecutionStore for TrackingExecutionStore {
 struct UnusedAdmissions;
 
 impl ClientAdmissionPort for UnusedAdmissions {
+    fn abandon(
+        &self,
+        key: &gateway_core::policy::ClientApiKeyId,
+        request: &gateway_core::engine::ModelRequestId,
+    ) {
+        let _ = futures::FutureExt::now_or_never(self.release(key, request));
+    }
+
     fn admit(
         &self,
         _: ClientAdmissionRequest,
@@ -1902,3 +1940,301 @@ fn start_operation_for_model(model: &str) -> Operation {
             .expect("OpenAI payload"),
     ))
 }
+
+#[derive(Default)]
+struct BoundedAdmissions {
+    active: Mutex<BTreeSet<ModelRequestId>>,
+    granted: AtomicUsize,
+    rpm_exhausted: AtomicBool,
+    hold_acquisition: AtomicBool,
+}
+
+#[derive(Default)]
+struct QueuedAccountProvider {
+    waiting: gateway_core::concurrency::ConcurrencyWaitQueue<&'static str>,
+}
+
+#[async_trait]
+impl Provider for QueuedAccountProvider {
+    fn name(&self) -> &'static str {
+        "openai"
+    }
+
+    fn catalog_generation(&self) -> ProviderCatalogGeneration {
+        ProviderCatalogGeneration::default()
+    }
+
+    async fn query_model_capabilities(
+        &self,
+    ) -> Result<Vec<ProviderModelCapabilities>, ProviderError> {
+        Ok(Vec::new())
+    }
+
+    async fn execute(
+        &self,
+        _: ProviderRequest,
+        context: AttemptContext,
+    ) -> Result<ProviderStream, ProviderError> {
+        let mut waiting = gateway_core::concurrency::CapacityWait::new(
+            &self.waiting,
+            gateway_core::concurrency::ConcurrencyQueuePolicy {
+                max_waiting: 1,
+                timeout: Duration::from_secs(5),
+            },
+            context.deadline(),
+            context.concurrency_wait_budget(),
+        );
+        loop {
+            waiting.wait(&["account"]).await.map_err(|error| {
+                ProviderError::new(error.provider_kind(), UpstreamSendState::NotSent)
+            })?;
+        }
+    }
+}
+
+#[test]
+fn account_wait_inherits_the_budget_spent_during_client_admission() {
+    use futures::FutureExt;
+    block_on(async {
+        let admissions = Arc::new(BoundedAdmissions::default());
+        let snapshot = start_snapshot_with_policy(
+            1,
+            true,
+            RateLimits {
+                max_concurrency: 1,
+                requests_per_minute: 0,
+            },
+        )
+        .with_client_queue_policy(gateway_core::concurrency::ConcurrencyQueuePolicy {
+            max_waiting: 1,
+            timeout: Duration::from_secs(1),
+        });
+        let service = DefaultExecutionService::new(
+            RuntimeSnapshotHandle::new(snapshot),
+            Arc::new(TrackingExecutionStore::default()),
+            ProviderRegistry::new(
+                [Arc::new(QueuedAccountProvider::default()) as Arc<dyn Provider>],
+            )
+            .unwrap(),
+            admissions.clone(),
+            Arc::new(UnusedCircuits),
+            Arc::new(UnusedContinuation),
+            Arc::new(RecordingClientApiKeyUsage::default()),
+        );
+        let running = service
+            .start(request(&service, ClientTransport::HttpSse))
+            .await
+            .unwrap();
+        let started_at = Instant::now();
+        let mut queued = service.start(request(&service, ClientTransport::HttpSse));
+        assert!(queued.as_mut().now_or_never().is_none());
+        futures_timer::Delay::new(Duration::from_millis(200)).await;
+        drop(running);
+        let mut started = queued.await.unwrap();
+        let error = started.session.collect_uncommitted().await.unwrap_err();
+        assert!(
+            matches!(
+                error,
+                EngineError::Provider(ref error)
+                    if error.kind() == ProviderErrorKind::ConcurrencyQueueTimeout
+            ),
+            "{error:?}"
+        );
+        assert!(
+            started_at.elapsed() < Duration::from_secs(3),
+            "账号层不能重新获得自己的 5 秒预算，实际耗时 {:?}",
+            started_at.elapsed()
+        );
+        assert!(admissions.active.lock().unwrap().is_empty());
+    });
+}
+
+impl ClientAdmissionPort for BoundedAdmissions {
+    fn abandon(&self, _: &ClientApiKeyId, id: &ModelRequestId) {
+        self.active.lock().unwrap().remove(id);
+    }
+
+    fn admit(
+        &self,
+        request: ClientAdmissionRequest,
+    ) -> BoxFuture<'_, Result<ClientAdmissionDecision, ClientAdmissionError>> {
+        Box::pin(async move {
+            use gateway_core::engine::admission::ClientAdmissionRejection;
+            if self.rpm_exhausted.load(Ordering::SeqCst) {
+                return Ok(ClientAdmissionDecision::Rejected(
+                    ClientAdmissionRejection::RateLimited,
+                ));
+            }
+            {
+                let mut active = self.active.lock().unwrap();
+                if !request.allow_concurrency_acquire
+                    || (request.limits.max_concurrency > 0
+                        && active.len() as u64 >= request.limits.max_concurrency)
+                {
+                    return Ok(ClientAdmissionDecision::Rejected(
+                        ClientAdmissionRejection::ConcurrencyLimited,
+                    ));
+                }
+                active.insert(request.model_request_id);
+            }
+            self.granted.fetch_add(1, Ordering::SeqCst);
+            if self.hold_acquisition.load(Ordering::SeqCst) {
+                futures::future::pending::<()>().await;
+            }
+            Ok(ClientAdmissionDecision::Granted)
+        })
+    }
+
+    fn release<'a>(
+        &'a self,
+        _: &'a ClientApiKeyId,
+        id: &'a ModelRequestId,
+    ) -> BoxFuture<'a, Result<bool, ClientAdmissionError>> {
+        Box::pin(async move { Ok(self.active.lock().unwrap().remove(id)) })
+    }
+
+    fn restore(
+        &self,
+        _: ClientAdmissionRecovery,
+    ) -> BoxFuture<'_, Result<ClientAdmissionRestoreResult, ClientAdmissionError>> {
+        Box::pin(async { Ok(ClientAdmissionRestoreResult::default()) })
+    }
+}
+
+fn queue_service(
+    max_concurrency: u64,
+    max_waiting: u32,
+    timeout: Duration,
+) -> (DefaultExecutionService, Arc<BoundedAdmissions>) {
+    let admissions = Arc::new(BoundedAdmissions::default());
+    let snapshot = start_snapshot_with_policy(
+        1,
+        true,
+        RateLimits {
+            max_concurrency,
+            requests_per_minute: 0,
+        },
+    )
+    .with_client_queue_policy(gateway_core::concurrency::ConcurrencyQueuePolicy {
+        max_waiting,
+        timeout,
+    });
+    let service = DefaultExecutionService::new(
+        RuntimeSnapshotHandle::new(snapshot),
+        Arc::new(TrackingExecutionStore::default()),
+        ProviderRegistry::default(),
+        admissions.clone(),
+        Arc::new(UnusedCircuits),
+        Arc::new(UnusedContinuation),
+        Arc::new(RecordingClientApiKeyUsage::default()),
+    );
+    (service, admissions)
+}
+
+#[test]
+fn client_queue_allows_five_running_five_waiting_and_rejects_the_eleventh() {
+    use futures::FutureExt;
+    block_on(async {
+        let (service, admissions) = queue_service(5, 5, Duration::from_secs(2));
+        let mut running = Vec::new();
+        for _ in 0..5 {
+            running.push(
+                service
+                    .start(request(&service, ClientTransport::HttpSse))
+                    .await
+                    .unwrap(),
+            );
+        }
+        let mut waiting = Vec::new();
+        for _ in 0..5 {
+            let mut future = service.start(request(&service, ClientTransport::HttpSse));
+            assert!(future.as_mut().now_or_never().is_none());
+            waiting.push(future);
+        }
+        let rejected = service
+            .start(request(&service, ClientTransport::HttpSse))
+            .await
+            .err()
+            .unwrap();
+        assert_eq!(rejected.kind(), GatewayErrorKind::ConcurrencyQueueFull);
+        assert_eq!(
+            admissions.granted.load(Ordering::SeqCst),
+            5,
+            "waiting must not consume admission/RPM"
+        );
+        drop(running.remove(0));
+        let first = waiting.remove(0).await.unwrap();
+        assert_eq!(admissions.active.lock().unwrap().len(), 5);
+        assert_eq!(admissions.granted.load(Ordering::SeqCst), 6);
+        assert!(waiting[0].as_mut().now_or_never().is_none());
+        drop((first, waiting, running));
+        assert!(admissions.active.lock().unwrap().is_empty());
+    });
+}
+
+#[test]
+fn cancelled_waiter_releases_its_place_and_new_requests_do_not_overtake_fifo() {
+    use futures::FutureExt;
+    block_on(async {
+        let (service, admissions) = queue_service(1, 2, Duration::from_secs(2));
+        let running = service
+            .start(request(&service, ClientTransport::WebSocket))
+            .await
+            .unwrap();
+        let mut first = service.start(request(&service, ClientTransport::WebSocket));
+        let mut cancelled = service.start(request(&service, ClientTransport::WebSocket));
+        assert!(first.as_mut().now_or_never().is_none());
+        assert!(cancelled.as_mut().now_or_never().is_none());
+        drop(cancelled);
+        drop(running);
+        let mut later = service.start(request(&service, ClientTransport::WebSocket));
+        assert!(
+            later.as_mut().now_or_never().is_none(),
+            "a newly freed slot belongs to the existing queue head"
+        );
+        let first = first.await.unwrap();
+        assert_eq!(admissions.granted.load(Ordering::SeqCst), 2);
+        drop(first);
+        let later = later.await.unwrap();
+        drop(later);
+        assert!(admissions.active.lock().unwrap().is_empty());
+    });
+}
+
+#[test]
+fn queue_timeout_and_rpm_rejection_leave_no_new_admission() {
+    block_on(async {
+        let (service, admissions) = queue_service(1, 1, Duration::from_millis(20));
+        let running = service
+            .start(request(&service, ClientTransport::HttpJson))
+            .await
+            .unwrap();
+        let timeout = service
+            .start(request(&service, ClientTransport::HttpJson))
+            .await
+            .err()
+            .unwrap();
+        assert_eq!(timeout.kind(), GatewayErrorKind::ConcurrencyQueueTimeout);
+        admissions.rpm_exhausted.store(true, Ordering::SeqCst);
+        let limited = service
+            .start(request(&service, ClientTransport::HttpJson))
+            .await
+            .err()
+            .unwrap();
+        assert_eq!(limited.kind(), GatewayErrorKind::RateLimited);
+        assert_eq!(admissions.granted.load(Ordering::SeqCst), 1);
+        drop(running);
+    });
+}
+
+#[test]
+fn cancelling_pending_admission_cleans_up_a_slot_acquired_before_the_reply() {
+    use futures::FutureExt;
+    let (service, admissions) = queue_service(1, 1, Duration::from_secs(2));
+    admissions.hold_acquisition.store(true, Ordering::SeqCst);
+    let mut pending = service.start(request(&service, ClientTransport::HttpSse));
+    assert!(pending.as_mut().now_or_never().is_none());
+    assert_eq!(admissions.active.lock().unwrap().len(), 1);
+    drop(pending);
+    assert!(admissions.active.lock().unwrap().is_empty());
+}

backend/crates/gateway-core/tests/main.rs

@@ -1,4 +1,5 @@
 mod account;
+mod concurrency;
 mod diagnostics;
 mod engine;
 mod error;

backend/crates/gateway-core/tests/routing/mod.rs

@@ -654,3 +654,60 @@ fn alias_should_only_be_available_from_a_provider_with_its_mapped_model() {
         )
     );
 }
+
+#[test]
+fn account_model_access_filters_catalog_and_aliases_using_the_whole_frozen_pool() {
+    use gateway_core::account::{AccountModelAccess, AccountModelAccessMode};
+    let provider = ProviderKind::new("openai").expect("provider");
+    let restricted =
+        AccountModelAccess::new(AccountModelAccessMode::Denylist, vec!["gpt-5.5".to_owned()])
+            .expect("policy");
+    let directory = Arc::new(RuntimeAccountDirectory::new(BTreeMap::from([(
+        ProviderAccountId::new("acct_restricted").expect("id"),
+        RuntimeAccount::new(provider.clone(), BTreeSet::new())
+            .with_model_access(restricted.clone()),
+    )])));
+    let old_scope = FrozenAccountScope::new(directory, ClientRoutingScope::all_accounts());
+    let snapshot = snapshot();
+    assert_eq!(
+        snapshot
+            .public_models_for_scope(&old_scope)
+            .iter()
+            .map(PublicModelId::as_str)
+            .collect::<Vec<_>>(),
+        vec!["grok-latest"]
+    );
+    assert!(!snapshot.contains_public_model_for_scope(
+        &PublicModelId::new("gpt-5.4").expect("alias"),
+        &old_scope
+    ));
+    let new_scope = FrozenAccountScope::new(
+        Arc::new(RuntimeAccountDirectory::new(BTreeMap::from([
+            (
+                ProviderAccountId::new("acct_restricted").expect("id"),
+                RuntimeAccount::new(provider.clone(), BTreeSet::new())
+                    .with_model_access(restricted),
+            ),
+            (
+                ProviderAccountId::new("acct_unrestricted").expect("id"),
+                RuntimeAccount::new(provider, BTreeSet::new()),
+            ),
+        ]))),
+        ClientRoutingScope::all_accounts(),
+    );
+    assert_eq!(
+        snapshot
+            .public_models_for_scope(&new_scope)
+            .iter()
+            .map(PublicModelId::as_str)
+            .collect::<Vec<_>>(),
+        vec!["gpt-5.4", "gpt-5.5", "grok-latest"]
+    );
+    assert!(
+        !snapshot.contains_public_model_for_scope(
+            &PublicModelId::new("gpt-5.5").expect("model"),
+            &old_scope
+        ),
+        "a new policy must not alter an existing request scope"
+    );
+}

backend/crates/gateway-store/src/admin_adapter.rs

@@ -2,10 +2,11 @@
 
 use super::*;
 
-pub(crate) struct AdminAuthStoreAdapter {
+pub(crate) struct AuthStoreAdapter {
     pub(crate) security: postgres::PgAdminSecurityAuditRepository,
     pub(crate) settings: postgres::PgRuntimeSettingsRepository,
-    pub(crate) state: redis::RedisAdminAuthStateRepository,
+    pub(crate) state: redis::RedisAuthStateRepository,
+    pub(crate) keys: postgres::PgAdminClientKeyStore,
 }
 
 pub(crate) struct AdminSettingsStoreAdapter {
@@ -43,6 +44,9 @@ impl SettingsStore for AdminSettingsStoreAdapter {
                 refresh_concurrency: command.refresh_concurrency,
                 max_concurrent_per_account: command.max_concurrent_per_account,
                 request_interval_ms: command.request_interval_ms,
+                max_waiting_per_key: command.max_waiting_per_key,
+                max_waiting_per_account: command.max_waiting_per_account,
+                concurrency_wait_timeout_seconds: command.concurrency_wait_timeout_seconds,
                 rotation_strategy: command.rotation_strategy.as_str().to_owned(),
                 model_mappings: store_model_mappings(command.model_mappings),
                 min_codex_desktop_version: command.min_codex_desktop_version,
@@ -62,6 +66,9 @@ impl SettingsStore for AdminSettingsStoreAdapter {
                     "refresh_concurrency".to_owned(),
                     "max_concurrent_per_account".to_owned(),
                     "request_interval_ms".to_owned(),
+                    "max_waiting_per_key".to_owned(),
+                    "max_waiting_per_account".to_owned(),
+                    "concurrency_wait_timeout_seconds".to_owned(),
                     "rotation_strategy".to_owned(),
                     "min_codex_desktop_version".to_owned(),
                     "min_codex_cli_version".to_owned(),
@@ -165,6 +172,9 @@ pub(crate) fn admin_runtime_settings(
         refresh_concurrency: settings.refresh_concurrency,
         max_concurrent_per_account: settings.max_concurrent_per_account,
         request_interval_ms: settings.request_interval_ms,
+        max_waiting_per_key: settings.max_waiting_per_key,
+        max_waiting_per_account: settings.max_waiting_per_account,
+        concurrency_wait_timeout_seconds: settings.concurrency_wait_timeout_seconds,
         rotation_strategy,
         min_codex_desktop_version: settings.min_codex_desktop_version,
         min_codex_cli_version: settings.min_codex_cli_version,
@@ -185,7 +195,7 @@ pub(crate) fn store_model_mappings(
 }
 
 #[async_trait::async_trait]
-impl AuthStore for AdminAuthStoreAdapter {
+impl AuthStore for AuthStoreAdapter {
     async fn load_password_hash(&self, admin_user_id: &str) -> AdminStoreResult<Option<String>> {
         postgres::AdminSecurityAuditRepository::password_hash(&self.security, admin_user_id)
             .await
@@ -213,45 +223,66 @@ impl AuthStore for AdminAuthStoreAdapter {
             .map_err(|error| admin_store_error("admin API key", error))
     }
 
-    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
-        redis::AdminAuthStateRepository::load_admin_session(&self.state, session_id)
+    async fn load_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        redis::AuthStateRepository::load_session(&self.state, session_id)
             .await
-            .map(|session| {
-                session.map(|record| AdminSession {
-                    admin_user_id: record.admin_user_id,
-                    expires_at: record.expires_at,
-                })
-            })
-            .map_err(|error| admin_store_error("admin session", error))
+            .map_err(|error| admin_store_error("authentication session", error))?
+            .map(auth_session)
+            .transpose()
     }
 
-    async fn store_session(
-        &self,
-        session_id: &str,
-        session: &AdminSession,
-    ) -> AdminStoreResult<()> {
-        redis::AdminAuthStateRepository::store_admin_session(
+    async fn store_session(&self, session_id: &str, session: &AuthSession) -> AdminStoreResult<()> {
+        let subject = match &session.subject {
+            SessionSubject::Admin { admin_user_id } => redis::SessionSubjectRecord::Admin {
+                admin_user_id: admin_user_id.clone(),
+            },
+            SessionSubject::Key { client_key_id } => redis::SessionSubjectRecord::Key {
+                client_key_id: client_key_id.as_str().to_owned(),
+            },
+        };
+        redis::AuthStateRepository::store_session(
             &self.state,
             session_id,
-            &redis::AdminSessionRecord {
-                admin_user_id: session.admin_user_id.clone(),
+            &redis::AuthSessionRecord {
+                subject,
                 expires_at: session.expires_at,
             },
         )
         .await
-        .map_err(|error| admin_store_error("admin session", error))
+        .map_err(|error| admin_store_error("authentication session", error))
     }
 
-    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AdminSession>> {
-        redis::AdminAuthStateRepository::delete_admin_session(&self.state, session_id)
+    async fn delete_session(&self, session_id: &str) -> AdminStoreResult<Option<AuthSession>> {
+        redis::AuthStateRepository::delete_session(&self.state, session_id)
             .await
-            .map(|session| {
-                session.map(|record| AdminSession {
-                    admin_user_id: record.admin_user_id,
-                    expires_at: record.expires_at,
-                })
-            })
-            .map_err(|error| admin_store_error("admin session", error))
+            .map_err(|error| admin_store_error("authentication session", error))?
+            .map(auth_session)
+            .transpose()
+    }
+
+    async fn client_key_enabled(
+        &self,
+        id: &gateway_core::policy::ClientApiKeyId,
+    ) -> AdminStoreResult<bool> {
+        self.keys.is_enabled(id).await
+    }
+
+    async fn consume_login_attempt(
+        &self,
+        source_ip: std::net::IpAddr,
+        source_limit: u32,
+        global_limit: u32,
+        window: std::time::Duration,
+    ) -> AdminStoreResult<Option<std::time::Duration>> {
+        redis::AuthStateRepository::consume_login_attempt(
+            &self.state,
+            &source_ip.to_string(),
+            source_limit,
+            global_limit,
+            window,
+        )
+        .await
+        .map_err(|error| admin_store_error("login limit", error))
     }
 
     async fn append_audit_event(&self, event: AdminAuditModel) -> AdminStoreResult<()> {
@@ -300,3 +331,26 @@ impl AuthStore for AdminAuthStoreAdapter {
         .map_err(|error| admin_store_error("admin audit", error))
     }
 }
+
+fn auth_session(record: redis::AuthSessionRecord) -> AdminStoreResult<AuthSession> {
+    let subject = match record.subject {
+        redis::SessionSubjectRecord::Admin { admin_user_id } => {
+            SessionSubject::Admin { admin_user_id }
+        }
+        redis::SessionSubjectRecord::Key { client_key_id } => SessionSubject::Key {
+            client_key_id: gateway_core::policy::ClientApiKeyId::new(client_key_id).map_err(
+                |_| {
+                    AdminStoreError::new(
+                        AdminStoreErrorKind::Invalid,
+                        "authentication session",
+                        "client key ID is invalid",
+                    )
+                },
+            )?,
+        },
+    };
+    Ok(AuthSession {
+        subject,
+        expires_at: record.expires_at,
+    })
+}

backend/crates/gateway-store/src/bundle.rs

@@ -108,13 +108,11 @@ pub async fn initialize(mut config: StoreConfig) -> StoreResult<StoreBundle> {
             Arc::new(postgres::PgAccountGroupRepository::new(pool.clone())),
             Arc::new(postgres::PgProxyRepository::new(pool.clone())),
         ),
-        Arc::new(AdminAuthStoreAdapter {
+        Arc::new(AuthStoreAdapter {
+            keys: postgres::PgAdminClientKeyStore::new(pool.clone()),
             security: postgres::PgAdminSecurityAuditRepository::new(pool.clone()),
             settings: postgres::PgRuntimeSettingsRepository::new(pool.clone()),
-            state: redis::RedisAdminAuthStateRepository::new(
-                redis_connection.clone(),
-                REDIS_NAMESPACE,
-            )?,
+            state: redis::RedisAuthStateRepository::new(redis_connection.clone(), REDIS_NAMESPACE)?,
         }),
         Arc::new(postgres::PgAdminClientKeyStore::new(pool.clone())),
         Arc::new(postgres::PgAdminObservabilityStore::new(

backend/crates/gateway-store/src/lib.rs

@@ -7,7 +7,7 @@ use std::sync::Arc;
 use std::time::{Duration, SystemTime};
 use std::{fmt, num::NonZeroU64, str::FromStr};
 
-use gateway_admin::model::auth::{AdminAuditEvent as AdminAuditModel, AdminSession};
+use gateway_admin::model::auth::{AdminAuditEvent as AdminAuditModel, AuthSession, SessionSubject};
 use gateway_admin::model::settings::{
     AdminApiKey, AdminApiKeyMutation, ModelMappings, ReplaceRuntimeSettings,
     RotationStrategy as AdminRotationStrategy, RuntimeSettings as AdminRuntimeSettings,

backend/crates/gateway-store/src/postgres/account_groups.rs

@@ -165,12 +165,12 @@ impl AccountGroupStore for PgAccountGroupRepository {
             .collect::<Vec<_>>();
         let rows = sqlx::query(
             "select membership.account_group_id,
-                    account.id, account.provider_kind, account.name, account.email,
+                    account.id, account.provider_kind, account.name, account.notes, account.email,
                     account.upstream_user_id, account.upstream_account_id, account.plan_type,
                     account.authentication_kind, account.credential_revision, account.outbound_proxy_url,
                     account.has_refresh_token, account.access_token_expires_at,
                     account.next_refresh_at, account.enabled, account.concurrency_limit,
-                    account.weight, account.credential_state, account.quota_access_state,
+                    account.weight, account.model_access_json, account.credential_state, account.quota_access_state,
                     account.quota_evidence, account.quota_access_observed_at,
                     account.quota_reset_at, account.last_error_reason,
                     account.last_error_message, account.credential_observed_at,