Add provider catalog runtime support

Cargo.lock

@@ -89,7 +89,10 @@ dependencies = [
 name = "anyllm_providers"
 version = "0.9.6"
 dependencies = [
+ "reqwest",
  "serde",
+ "serde_json",
+ "tokio",
 ]
 
 [[package]]

crates/providers/Cargo.toml

@@ -6,5 +6,15 @@ edition.workspace = true
 license.workspace = true
 repository.workspace = true
 
+[features]
+default = []
+runtime-catalog = ["dep:serde_json"]
+remote-catalog = ["runtime-catalog", "dep:reqwest"]
+
 [dependencies]
 serde = { version = "1", features = ["derive"] }
+serde_json = { version = "1", optional = true }
+reqwest = { version = "0.12", optional = true, default-features = false, features = ["native-tls", "stream"] }
+
+[dev-dependencies]
+tokio = { version = "1", features = ["macros", "rt-multi-thread"] }

crates/providers/README.md

@@ -98,9 +98,48 @@ Other public items (re-exported from the crate root):
 - Single-model lookup: `get_model(provider_id, model_id)`.
 - Migration helper: `canonical_provider_id(provider_id)`.
 
+## Runtime LiteLLM updates
+
+The default crate remains static and does no I/O. If your app needs newer
+LiteLLM provider/model rows without waiting for an `anyllm_providers` release,
+enable the opt-in runtime catalog features:
+
+```toml
+[dependencies]
+anyllm_providers = { version = "0.9", features = ["remote-catalog"] }
+```
+
+`runtime-catalog` adds owned catalog types and a parser for LiteLLM's
+`model_prices_and_context_window.json`. `remote-catalog` also adds explicit
+fetch/cache helpers using a caller-provided `reqwest::Client`:
+
+```rust
+use anyllm_providers::{ProviderCatalog, RemoteCatalogOptions};
+
+let client = reqwest::Client::builder()
+    .redirect(reqwest::redirect::Policy::none())
+    .timeout(std::time::Duration::from_secs(30))
+    .build()?;
+
+let cache_dir = std::env::temp_dir().join("anyllm-provider-catalog");
+let options = RemoteCatalogOptions::default()
+    .with_cache_dir(&cache_dir)
+    .with_stale_on_error(true);
+
+let catalog = ProviderCatalog::fetch_litellm_with_options(&client, &options).await?;
+for p in catalog.all_providers() {
+    println!("{}: {} models", p.id, catalog.list_models(&p.id).len());
+}
+```
+
+Known providers keep the bundled auth, protocol, env-var, and base-url metadata.
+Brand-new LiteLLM providers are exposed with `ProviderStatus::Stub`, a guessed
+API-key env var, and an empty default base URL so callers do not silently route
+to the wrong endpoint.
+
 ## Integrating into a TUI (or any client app)
 
-This crate gives you the catalog. It does **not** ship an HTTP client, an async runtime, or a key store, so a TUI integrates it as a read-only data source and supplies its own transport (typically `reqwest`) and config (typically `std::env` or a config file).
+In its default mode this crate gives you the static catalog. It does **not** ship an HTTP client, an async runtime, or a key store, so a TUI integrates it as a read-only data source and supplies its own transport (typically `reqwest`) and config (typically `std::env` or a config file).
 
 The recommended local integration can use **Ollama** and **LM Studio** as built-in providers. Both are local, OpenAI-compatible, require no API key, and their catalog entries (`auth: AuthKind::None`, `protocol: ProviderProtocol::OpenAICompat`) reflect that. Ollama is part of the LiteLLM snapshot; LM Studio is a legacy-only local provider that still resolves through `get_provider("lm_studio")` but is not returned by `all_providers()`.
 

crates/providers/src/lib.rs

@@ -1,8 +1,16 @@
+#[cfg(feature = "runtime-catalog")]
+pub mod catalog;
 pub mod model;
 pub mod provider;
 pub mod providers;
 pub mod registry;
 
+#[cfg(feature = "runtime-catalog")]
+pub use catalog::{
+    CatalogError, CatalogMetadata, OwnedModelDef, OwnedProviderDef, ProviderCatalog,
+};
+#[cfg(feature = "remote-catalog")]
+pub use catalog::{RemoteCatalogOptions, DEFAULT_MAX_CATALOG_BYTES, LITELLM_CATALOG_URL};
 pub use model::{ModelCapabilities, ModelDef, ModelStatus};
 pub use provider::{AuthKind, ProviderCapabilities, ProviderDef, ProviderProtocol, ProviderStatus};
 pub use registry::{

crates/providers/src/model.rs

@@ -1,5 +1,5 @@
 /// Features a specific model supports.
-#[derive(Debug, Clone, Copy, Default)]
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub struct ModelCapabilities {
     pub streaming: bool,
     pub tool_use: bool,
@@ -9,7 +9,7 @@ pub struct ModelCapabilities {
 }
 
 /// Availability of a model.
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum ModelStatus {
     Available,
     Deprecated,

crates/providers/src/provider.rs

@@ -1,6 +1,6 @@
 /// Wire format / HTTP client strategy used to communicate with a provider.
 /// Maps to proxy's `BackendKind` variants.
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum ProviderProtocol {
     /// Standard OpenAI Chat Completions (Groq, Together, Mistral, etc.)
     OpenAICompat,
@@ -21,7 +21,7 @@ pub enum ProviderProtocol {
 }
 
 /// How the provider expects authentication.
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum AuthKind {
     /// `Authorization: Bearer <key>`
     Bearer,
@@ -36,7 +36,7 @@ pub enum AuthKind {
 }
 
 /// Implementation maturity of a provider.
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum ProviderStatus {
     /// HTTP client exists and has been live-tested.
     Implemented,
@@ -48,7 +48,7 @@ pub enum ProviderStatus {
 }
 
 /// Capabilities advertised for a provider (endpoint-level, not model-level).
-#[derive(Debug, Clone, Copy, Default)]
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub struct ProviderCapabilities {
     pub chat_completions: bool,
     pub streaming: bool,

crates/providers/src/registry.rs

@@ -93,6 +93,26 @@ static LEGACY_ONLY_MODELS: &[(&str, &[ModelDef])] = &[
     ("xinference", providers::xinference::MODELS),
 ];
 
+#[cfg(feature = "runtime-catalog")]
+pub(crate) fn advertised_provider_defs() -> &'static [&'static ProviderDef] {
+    ALL_PROVIDERS
+}
+
+#[cfg(feature = "runtime-catalog")]
+pub(crate) fn legacy_only_provider_defs() -> &'static [&'static ProviderDef] {
+    LEGACY_ONLY_PROVIDERS
+}
+
+#[cfg(feature = "runtime-catalog")]
+pub(crate) fn advertised_model_groups() -> &'static [(&'static str, &'static [ModelDef])] {
+    ALL_MODELS
+}
+
+#[cfg(feature = "runtime-catalog")]
+pub(crate) fn legacy_only_model_groups() -> &'static [(&'static str, &'static [ModelDef])] {
+    LEGACY_ONLY_MODELS
+}
+
 /// Return the LiteLLM-canonical provider id for a local legacy provider id.
 pub fn canonical_provider_id(id: &str) -> &str {
     PROVIDER_ALIASES

crates/proxy/Cargo.toml

@@ -10,7 +10,7 @@ repository.workspace = true
 anyllm_translate = { path = "../translator", version = "0.9.6" }
 anyllm_client = { path = "../client", version = "0.9.6" }
 anyllm_batch_engine = { path = "../batch_engine", version = "0.9.6" }
-anyllm_providers = { path = "../providers", version = "0.9.6" }
+anyllm_providers = { path = "../providers", version = "0.9.6", features = ["runtime-catalog"] }
 axum = { version = "0.8", features = ["ws", "multipart"] }
 tokio = { version = "1", features = ["full"] }
 reqwest = { version = "0.12", default-features = false, features = ["json", "stream", "native-tls", "http2", "multipart"] }