Brennan Benson 18bbf6f209 refactor(orchestration): resolve every caller and target to one orchestration party, keyed by the Orca session id (#22555)
* refactor(orchestration): resolve a session caller at the dispatch entry and bind it by actor

WIP: entry resolver on both dispatchers, caller identity through run scope,
actor-keyed Run binding and unbind, actor writes on bind/create/assign, and
actor-aware mail ownership exclusions.

* test(orchestration): pin actor-keyed Run binding, stale-actor precedence and actor mail ownership

Keeps the non-session dispatch path synchronous so terminal and session-tab
streams reach their handler without an extra async hop.

* feat(orchestration): resolve session callers before params parse and pin every verb on both routes

A session caller need not name itself in a param that requires a caller: the
entry binds the declared caller to the session before the schema runs. Session
refusal codes pass through the RPC error map, and DB row reads added here carry
their SAFETY rationale.

* test(orchestration): pin the SSH check's pane through the caller-identity lookup

* test(orchestration): pin a Run-less session's direct check and receipt binding without a caller param

Drops the actor clause from self-dispatch detection: a creator and assignee can
only share an actor when they already share a handle or pane.

* test(orchestration): pin a session's own Run for plain and group sends and the assignee-only mail sweep

* test(orchestration): name the party-naming field population for its role

* fix(orchestration): clear a worker actor an older binary's unbind leaves, and refuse a worker without its identity

An older binary unbinds a structured worker's Run by clearing handle and pane, which
leaves the actor looking like a handle-less chat binding. The every-open repair
clears that shape for actors recorded as structured workers only, and the resolver
refuses a worker session whose worker identity is gone, so this binary never
writes the shape itself.

* refactor(orchestration): read a Run's coordinator actor through its generation, and give a party's addresses one owner

The coordinator actor now counts only at the consumer generation it was written at, so a
Run binding matches a session by that rule alone. It replaces two mechanisms for the same
fact: the rule that an actor beside a handle it did not bind with never matches, and the
open-time repair that cleared a structured worker's actor an older binary's unbind left.
Every write of an older binary that rebinds or unbinds bumps the generation, so both
shapes stop counting by themselves, including a chat's actor after a rebind then an
unbind, which neither old mechanism caught. createRun and the same-coordinator actor
correction write the generation in the statement that writes the actor. The resolver
still refuses a structured worker whose worker identity is gone; its predicate moves
next to the worker identity lookup.

addressSpellingsOf is the one owner of the addresses a party is reachable at (a
structured worker's handle and session actor). createRun, bindRun, the coordinator
unbind and the declared-caller check use it instead of hand-built sets, and each has a
test at both of a worker's addresses.

* fix(orchestration): say a released session is not running, and scope the pane-key credential claim to requests without a session

A released lease is evicted, not ended: a user turn resumes the session, so the refusal
now says it is not running right now instead of that it has ended.

The worker pane-key comment claimed the random leaf is what stops anyone who learns a
session id from acting as the worker. On the same-host socket route the session id now
names the worker with no token by design; the pane key still matters where a request
names no session (a PTY agent's, or the paired-client route, which refuses session ids).

* test(orchestration): pin the same-coordinator actor correction's generation write on a row an older binary wrote

* refactor(orchestration): resolve session callers by the bare Orca session id, typed apart from its address

Carries the Orca session id rename into caller resolution, Run binding and Dispatch
creation. The caller identity holds the bare `orcaSessionId`; the `session:<id>`
spelling is derived by formatOrcaSessionAddress wherever mail needs it.

`OrcaSessionId` and `OrcaSessionAddress` are distinct branded strings. Only
isOrcaSessionId and parseOrcaSessionAddress produce an id, and only
formatOrcaSessionAddress produces an address, so comparing the two is a type error.
The Orca session id columns on the row types carry the id type.

Every reader that compares a stored id with a mail address now compares like with
like: the active-Dispatch ownership check formats the stored id
(orcaSessionAddressSql), and the stray-mail sweep, the creator nesting lookup and the
recorded-worker check bind a parsed or typed bare id. The Run-mailbox ownership
check keeps its existing handle comparison beside the session one.

* fix(orchestration): accept a worker's ask to every address its coordinator is reachable at

A worker's preamble names its coordinator as `session:<id>` when the coordinator is a structured session, but ask only accepted `run:<id>` or the coordinator's terminal handle. A chat coordinator has no handle, and a coordinating structured worker has two addresses, so ask --to the session address was refused as dispatch_run_mismatch. The check now takes the Run's current coordinator addresses from addressSpellingsOf(runCoordinatorKey(run)).

* refactor(orchestration): require every caller-identity entry point to be handed the resolved session

The resolved session parameter was optional on resolveRunScope, resolveOrchestrationCaller, orchestrationCallerIdentity, resolveDispatchCreator and resolveDispatchCallerWorktreeId, so a method that forgot to pass it would compile and silently treat a chat's session address as a terminal handle. It is now required and typed `OrchestrationSessionCaller | undefined`, so leaving it out is a type error. Every call site already passed it; no behavior changes.

* fix(orchestration): deliver mail sent to a session address to the mailbox that session reads

A send to session:<id> was resolved like a terminal handle: no live pane, so a chat's
current Run was missed (two Runs read as ambiguous), a Run-less chat was refused though
it reads its direct mailbox, and a structured worker's session address never reached
its Dispatch. Resolve a worker's session address as its handle, a chat's by its bound
Run, and fall back to the chat's durable direct mailbox while it runs on this host.

* refactor(orchestration): resolve every party through one resolver with one mailbox address

A structured worker is reachable at its handle and at its session address, and a chat only at its
session address. Callers and targets were each compared or resolved at their own site, some against
one spelling and some against every spelling, and the sites that did neither refused or misrouted.

Add orchestration-party: resolveOrchestrationParty (and resolveOrcaSessionParty for a bare id) is
now the only place an address becomes a party. The session caller, the declared-caller check,
ask's target and inbox's filter all resolve through it, so caller and recipient resolution cannot
disagree, including on a recorded worker whose identity this host lost. Every session-to-party step
passes through canonicalOrcaSessionId, the seam later lineage canonicalization plugs into.

mailboxAddressOf replaces addressSpellingsOf: a party has one mailbox address (a worker's handle
today), and Run binding remembers and reroutes only that.

A request with no session id that declares a session address as its caller now gets the party it
names: a worker's handle, as if it had named it, or a session_caller_chat_not_declarable refusal
for a chat, which is identified only by the session id its own environment sends.

* fix(orchestration): route dispatch and mail targets through the party they name

dispatch --to a worker's session address stored that address as the assignee handle, so the worker,
which reads by its handle, never saw the Dispatch. The assignee now resolves through the party
resolver: a worker's either spelling assigns its handle and records its Orca session id. A chat
cannot receive a dispatch yet, so dispatch --to a chat is refused with
session_chat_not_dispatchable before any row is written.

Recipient routing resolves the party once and, for any session-backed party, finds its Run by its
durable binding rather than by a live pane. A structured worker that coordinates a child Run while
assigned in its parent got mail at its parent Dispatch mailbox whenever its session was evicted,
which it never reads while bound to the child Run. Terminal handles keep the same live-pane lookup.

* refactor(orchestration): delete the SQL that matched a worker's second spelling

With every caller and target resolved to one mailbox address, no writer stores mail under a
structured worker's session address: sends resolve it to the handle, a declared session caller is
rewritten to the handle or refused, replies answer a stored canonical sender, and questions,
answers, escalations, federation and legacy mail write run:, dispatch: or handle addresses (legacy
rows are legacy_direct, which these queries never read). The branches that matched that spelling
were unreachable and are removed:

- the session-address OR in activeDispatchOwnsAddressSql, back to the assignee handle alone;
- the session branches in routeForeignDirectMessagesToOwnedMailboxes and
  findActiveDispatchForDirectMessageOwner, which return to the base-branch form.

Tests that inserted such rows directly now pin the behaviour through the verbs: both spellings of a
worker land in the one mailbox it reads.

* test(orchestration): pin one canonical address per party across every caller and target

- every target param in ORCHESTRATION_TARGET_PARAM, for a worker's handle and session address and
  a chat's session address: send, ask, dispatch (a chat refused with no row) and inbox;
- a declared session-address caller on a request with no session id: a worker acts as its handle,
  a chat is refused;
- a worker that coordinates a child Run while assigned in its parent gets mail in the child Run
  with no live pane;
- no mail writer stores to_handle or from_handle as a worker's session address;
- the caller a session id resolves to and the recipient its address resolves to agree in every
  party state, and every session-to-party step goes through canonicalOrcaSessionId.

* refactor(orchestration): drop the session caller's chat-to-terminal-view handoff remnants

The structured-chat terminal handoff is gone from the base branch: a session is owned natively only.

- the lease rule no longer speaks of either owner or a handoff keeping identity;
- an in-progress owner change (new-owner-proving, recovering, manual-recovery, which the base branch
  still produces) is refused as changing owners, not as switching between chat and terminal view;
- the native to terminal-view to native identity test is deleted, and the terminal-evidence test
  no longer describes that evidence as a terminal view's.

* fix(orchestration): read a declared caller param by typed access, not Reflect.get
2026-09-25 18:01:40 -07:00
2026-09-25 22:14:01 +00:00
2026-05-04 20:42:03 -07:00
2026-03-16 22:27:51 -07:00
2026-03-28 10:19:14 -07:00

Orca Orca

GitHub stars Total downloads across all releases License: MIT Join the Orca Discord Follow Orca on X Supported platforms: macOS, Windows, and Linux

中文 · 日本語 · 한국어 · Español · Français · Português

The AI Orchestrator for 100x builders.
Run Codex, ClaudeCode, OpenCode or Pi side-by-side — each in its own worktree, tracked in one place.

Download Orca

Orca desktop app running agents in parallel worktrees, with the Orca mobile companion app in the corner

Features

Mobile Companion

Monitor and steer your agents from your phone — get notified when an agent finishes and send follow-ups from anywhere.

iOS App Store · TestFlight · Android APK 0.0.50 · Docs →

Orca desktop with the mobile companion app

Parallel Worktrees

Fan one prompt across five agents, each in its own isolated git worktree — compare the results and merge the winner.

Docs →

Parallel worktree orchestration

Terminal Splits

Ghostty-class terminals with WebGL rendering, infinite splits, and scrollback that survives restarts.

Docs →

Terminal splits

Design Mode

Click any UI element in a real Chromium window to send its HTML, CSS, and a cropped screenshot straight into your agent's prompt.

Docs →

Embedded browser and Design Mode

GitHub & Linear, Native

Browse PRs, issues, and project boards in-app — open a worktree from any task and review without a context switch.

Docs →

GitHub and Linear task workflows in Orca

SSH Worktrees

Run agents on a beefy remote box with full file editing, git, and terminals — auto-reconnect and port forwarding included.

Docs →

Remote worktrees over SSH

Annotate AI Diffs

Drop comments on any diff line and ship them back to the agent — review, edit, and commit without leaving Orca.

Docs →

Annotate AI-generated diffs

Drag Files to Agents

VS Code's editor with autosave everywhere — drag files or images straight into an agent prompt.

Docs →

Drag files and images into an agent prompt

Orca CLI

Agents drive Orca too — script every workflow with orca worktree create, snapshot, click, and fill.

Docs →

Script Orca from the CLI

Also in the box:

  • Quick open — Search across worktrees, files, agents, commands, and repo context without leaving your flow.
  • Account switcher & usage tracking — See Claude and Codex usage and rate-limit resets, and hot-swap accounts without re-logging in.
  • Rich repo previews — Preview Markdown, images, PDFs, and repo docs in the workspace.
  • Computer Use — Let agents operate desktop apps and visible UI when a workflow needs real interaction.
  • Notifications and unread state — Know when an agent finishes or needs attention, then mark threads unread to come back later.
  • And many, many more — we ship daily, so this list is perpetually behind. The changelog is the real feature list.

Supported Agents

Works with any CLI agent — if it runs in a terminal, it runs in Orca.

Claude Code logo Claude Code   Codex logo Codex   Grok logo Grok   Cursor logo Cursor   GitHub Copilot logo GitHub Copilot   Muse logo Muse   ZCode logo ZCode   OpenCode logo OpenCode   MiMo Code logo MiMo Code   Amp logo Amp   OpenClaude logo OpenClaude   Antigravity logo Antigravity   Pi logo Pi   oh-my-pi logo oh-my-pi   Hermes Agent logo Hermes Agent   Devin logo Devin   Goose logo Goose   Auggie logo Auggie   Autohand Code logo Autohand Code   Charm logo Charm   Cline logo Cline   Codebuff logo Codebuff   Command Code logo Command Code   Continue logo Continue   Droid logo Droid   Kilocode logo Kilocode   Kimi logo Kimi   Kiro logo Kiro   Mistral Vibe logo Mistral Vibe   Qwen Code logo Qwen Code   Rovo Dev logo Rovo Dev   + any CLI agent


Install

Desktop — macOS, Windows, Linux

Or via a package manager:

# macOS (Homebrew)
brew install --cask stablyai/orca/orca

# Arch Linux (AUR) — or stably-orca-git to build from source
yay -S stably-orca-bin

Mobile Companion — iOS, Android

Pair with your desktop app to monitor and steer your agents from your phone.


Community & Support

  • Discord: Join the community on Discord.

  • Twitter / X: Follow @orca_build for updates and announcements.

  • WeChat: Scan to join the Orca community WeChat group 10.

    WeChat group 10 QR code for the Orca community
  • Feedback & Ideas: We ship fast. Missing something? Request a new feature.

  • Privacy: See the privacy & telemetry docs for what anonymous usage data Orca collects and how to opt out.

  • Show Support: Star this repo to follow along with our daily ships.


Developing

Want to contribute or run locally? See our CONTRIBUTING.md guide.

The relay that pairs the mobile app with a desktop host is also in this repository under cloud/, with a separate pnpm workspace and setup guide.

Orca contributors

GitHub star history chart for stablyai/orca

Signed Builds

Windows code signing sponored/provided by SignPath.io, certificate by SignPath Foundation.

License

Orca is free and open source under the MIT License.

S
Description
Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.
Readme MIT
1.8 GiB
Languages
TypeScript 95.4%
JavaScript 3.9%
Swift 0.2%
HCL 0.1%
CSS 0.1%