Agent Orchestrator
github.com/composiohq/agent-orchestrator- Category
- AI Agents
- Rank
- No. 242Tools index
Previous survey · No. 249 ·
- Pricing
- Open Source
- Platform
- cli · web
- Type
- TOOL
- Builder
- @ComposioHQ
- GitHub
- 12.1k stars
- Latest release
- v0.13.1-nightly.202609151742
- Date
About
An orchestration platform for managing fleets of parallel AI coding agents. Each agent works in its own git branch to autonomously handle CI fixes, merge conflicts, and code reviews while you supervise from a centralized dashboard.
What it does
Agent Orchestrator is a local agent IDE that gathers coding sessions, pull requests, browser previews, chat, and terminals into one control surface. It creates a separate worktree for each Git-backed session, while Scratch sessions use managed branchless directories. A local daemon tracks durable session facts and sends updates to desktop, mobile, and command-line clients.
Why it's ranked here
The architecture addresses the difficult parts of multi-agent development, not just process launching. It preserves workspace isolation, observes GitHub and runtime state, and returns failures or review feedback to the responsible session. Its broad adapter support and multiple control surfaces make it credible, though several reviewers remain explicitly experimental and host-trusted.
What's good
Session state survives beyond a terminal window because the daemon stores durable facts in SQLite and derives display status when clients read it. Chat and terminal modes share the same lifecycle boundaries. The product also supports typed API contracts, live event updates, pull request observation, authenticated mobile access, and isolated browser storage between workers.
Tradeoffs
This is a local, single-user control plane, not a distributed orchestration service. It depends on installed Git and agent command-line tools, plus terminal runtimes that differ by operating system. Some reviewer adapters are experimental and lack operating-system or network containment. Mobile contributor setup is also documented as incomplete.
How to use it well
Use it when one engineer regularly runs several coding agents against the same repository and needs clear ownership of branches, feedback, terminals, and pull requests. Start with the desktop app, then add command-line or mobile control where useful. It does not replace the coding agents themselves, provide their authentication, or serve as a distributed team platform.
Technical notes+
The architecture in docs/architecture.md centers on a Go daemon, SQLite facts, trigger-backed change capture, REST, SSE, and terminal WebSockets. docs/stack.md records chi, Cobra, sqlc, goose, tmux on Darwin/Linux, conpty on Windows, and Electron with TypeScript. package.json exposes Go linting, frontend type checks, SQL generation, and OpenAPI-to-TypeScript generation. docs/STATUS.md says the working loop covers project creation through pull-request merge, while docs/daemon-environment.md documents a proposed fix for GUI-launched processes inheriting an incomplete shell environment. packages/mobile/app/(tabs)/index.tsx shows an Expo dashboard with pairing, connection recovery, session grouping, notifications, and worker spawning.
Observed
- License
- Apache-2.0
- Languages and shells
- Go backend daemon; Electron and React desktop frontend in TypeScript; Expo and React Native mobile companion
- Install surface
- Desktop builds are provided for Apple silicon and Intel macOS, Windows x64, Linux AppImage, Debian/Ubuntu, and Fedora/RHEL
- Interfaces
- Desktop app, command-line interface, mobile companion, REST API, SSE event stream, and terminal WebSocket
- Storage and API structure
- Local SQLite storage in WAL mode, generated SQL access, SQL migrations, and OpenAPI-generated TypeScript types
- Testing structure
- Go tests, Vitest frontend unit tests, and Playwright renderer end-to-end tests are documented
Read from README.md, package.json, packages/mobile/app/(tabs)/index.tsx, docs/stack.md, docs/README.md, docs/STATUS.md, docs/telemetry.md, docs/development.md, docs/architecture.md, docs/daemon-environment.md.
What it can do
Deploy multiple AI coding agents in parallel
Codebase and configuration parameters → Fleet of active AI agents working on separate git branches
Automatically fix CI/CD pipeline failures
Failed CI/CD builds and error logs → Code fixes and updated commits
Resolve merge conflicts autonomously
Conflicting code changes between branches → Resolved merge conflicts and clean merges
Process and respond to code review comments
Code review feedback and suggestions → Code changes addressing review comments
Create and manage separate git worktrees for each agent
Git repository and agent assignments → Isolated git worktrees and branches for parallel work
Monitor agent activities from centralized dashboard
Agent status and work progress data → Real-time dashboard showing fleet status and activities
Handle pull request workflows automatically
Code changes and PR requirements → Created, updated, or merged pull requests
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.