OrcaSynth, a tool for managing and orchestrating autonomous AI coding agents, is now available as open-source under the MIT license. The full project is on GitHub and provides a complete solution for working with multiple AI agents at once — from planning and spawning to supervision and safety boundaries.
What OrcaSynth does
The core concept is autopilot planning. You give it a goal and OrcaSynth breaks it into phases, assigns an agent to each, and runs them in isolated tmux sessions. It supports Claude Code, OpenCode, and Codex — agent selection is configurable per task.
Each phase runs in its own session while the system monitors progress and automatically advances work based on dependencies. If an agent gets stuck, a built-in detector notices and offers a resolution.
Controlled autonomy
Not every task requires the same level of oversight. OrcaSynth works with autonomy levels ranging from full supervision to complete independence. Safety rules evaluate risk areas — such as database schema changes, authentication, or payment logic — before deploying an agent. Sensitive tasks stay paused until explicitly approved by a human.
Token usage in real time
One of the recent additions is real-time token usage display in the interface. Every session shows input tokens, cache hits, and output tokens with a detailed breakdown on hover. This gives you an immediate picture of what each agent consumed and where savings are possible.
Autopilot in the timeline
Autopilot mission work is now displayed in the timeline with clear labels. Each phase and the overall mission carry an Autopilot tag, so it is immediately obvious which part of the work is running automatically and which requires intervention.
Deployment without external dependencies
OrcaSynth is designed as a standalone daemon with an SQLite database. No external services required — just Node.js 22+, tmux, and access to an LLM provider. The web UI runs on Next.js, and the API is built on Hono.
Getting started is straightforward:
npm install && npm run build
node dist/daemon/index.jsThe web UI connects via REST API and SSE event stream. Everything runs locally, and your data stays on your machine.
Open code, open access
The project is under the MIT license, with full documentation, a test suite, and CI pipeline. Community contributions are welcome — whether bug fixes, new features, or documentation improvements.
For teams and individuals looking to integrate AI agents into their workflow without relying on cloud services, this is a practical path: download, run, adapt.
OrcaSynth is available on GitHub.