We Open-Sourced Timon: A Workstation for Coordinating AI Coding Agents
AI coding agents are getting super good at individual tasks, however anyone who has tried to use several of them on a real project knows the hard part isn't just generating code.
The complexity is coordinating the work and keep consistency.
Who breaks a feature into smaller tasks ? Who reviews the changes ? How do you keep two agents from editing the same files ? What happens when a model hits its quota, or a long-running task needs to resume later ? And who gets the final say before code is pushed or published ?
Those are the problems Timon is designed to address and Timon's source code is now public under AGPL-3.0 and MIT licenses.
From chat sessions to an agent team
Timon runs on your machine and works with supported coding-agent CLIs you already use, including Claude Code and Codex amongst others. Rather than managing each session independently, you work with an orchestrator that delegates tasks to specialized roles.
A typical workflow looks like this:
- Describe the goal: the orchestrator plans and dispatches tasks.
- Work isolation: coding tasks use separate Git work trees.
- Independent review: a reviewer using another model examines the diff.
- Merge deliberately: a merger runs checks before integrating the result.
- Keep the human involved: important actions still require explicit approval before moving on.
This isn't a promise that agents never make mistakes. It's a way to make their work more visible, reviewable and manageable.
Why local-first matters
Timon's daemon runs on Windows, macOS and Linux. It keeps mission state so you can close the client and come back later. Supported agent CLIs run using the subscriptions you already have, Timon does not silently resell model tokens.
The architecture also supports plugins, provider extensions, scheduled routines and other ways to tailor the workstation to your own workflows.
What open source means here
The repository is public and the core is licensed under AGPL-3.0 while the SDK and plugin templates are under MIT to allow developers to build and distribute their own plugins, including commercial ones. Please check the repository's LICENSE and LICENSING.md for the exact terms.
Open sourcing mattered to us because developers can inspect the implementation, report issues, contribute translations and build plugins that can benefit the whole community.
That scrutiny is especially important for tools that can execute code and interact with your repositories.
Try it yourself
- Source: https://github.com/ai-timon/timon
- Installation and documentation: https://docs.timon-ai.com/
- Website: https://timon-ai.com/
- Contributions: https://github.com/ai-timon/timon/blob/main/CONTRIBUTING.md
- Discord: https://timon-ai.com/go/discord
If you test it, we recommend to try with a small repository first. Give the orchestrator a feature with a clear acceptance test, inspect the resulting work trees and review the output. Then see whether the approval flow fits what you expect from such an orchestrator.
Our intention: collect feedback to make Timon better
For developers already using Claude Code, Codex or multiple AI agents:
- What is your biggest pain point when coordinating them?
- Would you rather let different models code and review, or keep one model throughout?
- What security or transparency requirements would you have before adopting an agent workstation?
The source is open, and critical feedback is welcome.
We'd rather hear what needs to improve than collect another generic 'looks awesome' comment.
Top comments (0)