Skip to content

Timon Puts a Team of AI Coding Agents on Your Machine

ai-timon/timon is an open-source agentic workstation that splits a job across coder, reviewer and merger agents and runs them through CLI subscriptions you already pay for.

Draftship Editorial4 min read

Timon, an open-source project published by ai-timon, describes itself as an agentic workstation: one place from which a person directs several AI coding agents. The repository had 619 stars and 79 forks on GitHub as of 2026-11-19, 13 days after it was created on 2026-11-06, and it is appearing in GitHub trending listings for AI projects. It is written in TypeScript and licensed AGPL-3.0, so the Timon agentic workstation is young and small, and stars measure interest rather than use.

What the source says it does

You describe a job in plain words, by voice if you prefer. An orchestrator breaks it into tasks and hands each one to an agent. Every project starts with six roles: an Orchestrator that plans and tracks blocked work, a Coder that works in its own git worktree and commits step by step, a Reviewer that reads the diff with a different model and sends it back when it is wrong, a Merger that runs the full test suite once and lands the branch, a Researcher that turns an idea into a brief, and an Ops agent for dependencies, clean-ups and release notes. The README says roles are fields rather than names, so they can be renamed, added to, or replaced with agents you already have.

Agents ask before they push, publish or delete anything, and secrets are kept where the agent does not see them. A rule learned in one mission is injected into the next, every turn and tool call is stored so a dead mission can be resumed, and when one subscription runs dry the work moves to the next backend. There is a built-in browser isolated per mission with annotations and recordings, plus routines, cron and webhook triggers.

The subscription model, and what you need installed

The pitch is that agents run through CLI subscriptions you already pay for, and the README states that Timon never calls a token-billed API on your behalf. You need one agent CLI already signed in: claude, codex, copilot, grok, opencode, pi or hermes-acp. Installation is a shell or PowerShell one-liner that picks the package for your system, downloads it from a signed release list, and checks its size and checksum before installing. The installer brings its own Node.js and adds Git on Windows when it is missing. Two commands, timon doctor and timon backends, check the service and list the agent CLIs that were found. On a server without a screen, timon service install runs the daemon as a systemd unit, launchd job or Windows scheduled task.

The daemon holds all state and the clients hold none, so closing the app does not stop the agents. Clients exist for desktop, browser, phone, Telegram, CLI and a remote API. The stack listed in the project's GitHub repository is TypeScript, Node.js 22, Cordis, SQLite, Next.js, React, Tailwind and Electron.

Free, Pro and what is not said

The free and Pro tiers run the same app with the same features. Pro adds several users with roles and groups, per-person permissions, and support; the README points to a plans page instead of naming a price. A licence never replaces a CLI subscription. Plugins, including third-party ones, are signed and installed without a restart, and are built against an SDK on npm. The licence is AGPL-3.0 with an exception that lets plugins use any licence, while the SDK and templates are MIT.

Limits and open questions

There are no benchmarks, no independent reviews and no stated user counts, so the code, review and merge loop is described but not measured. The Pro price is not given. Platform support is narrow in places: Windows 10+ on x64, macOS 11+ on Apple silicon only, Linux x64 with glibc 2.28 or newer. Whether third-party CLI subscriptions permit this kind of orchestration is not addressed. The README says plugins are signed but does not describe how third-party plugins are reviewed, and a first pull request requires signing a CLA.

What it means for building apps with AI

The interesting part is the shape of the workflow rather than the tool: separate agents for writing, reviewing with a different model and merging, each task isolated in its own worktree, with a human approval step before anything ships. That pattern holds whether the agents run locally or hosted. If you are building an app and want to control which model handles which step, Draftship lets you choose the AI model on paid plans. Draftship is not affiliated with ai-timon and does not support Timon.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.