← Back to Discover

Garcon

A self-hosted browser workspace for running several coding agents side by side, steering their work, reviewing diffs and committing from one place, with a CLI and a phone-friendly layout.

Coding agentsCode review

What it does

Garcon is a workspace you host yourself and open in a browser. It runs coding agents side by side, naming Claude Code, Codex, Cursor Agent, OpenCode, Amp, Factory Droid and Pi, next to terminals, files and Git, so a task can go from prompt to reviewed commit without leaving the page. While agents work you can queue and reorder instructions, steer a running turn, approve tool use, answer questions, or interrupt. Chats can be arranged on canvases or a tag-driven board, linked by fork, handoff and delegation lineage, and passed messages between chats with their origin labelled. Workspace tickets track work separately from chats. Besides the agents' own logins, it can connect to model endpoints compatible with the Anthropic Messages, OpenAI Responses and OpenAI Chat Completions formats, with presets for providers such as Ollama, OpenRouter and Gemini. A CLI drives the same chats from scripts, and remote executors can run agents, files and Git on other machines. Setup needs Bun, git and a modern browser, plus at least one working agent or API provider, and a Docker image is published. The page says agents and their commands run on the host under your account and are not sandboxed. Pull request commands need an authenticated GitHub CLI. It is licensed under GPL-3.0.

Capabilities

  • Runs several coding agents in parallel in one browser workspace
  • Built-in terminals, file browser and editor, diff review and line-level staging
  • Steering controls: queue instructions, approve tools, interrupt
  • Canvases, boards and tickets for organising parallel work
  • CLI and remote executors for scripted or multi-machine setups
  • Responsive layout for use from a phone

When to use it

  • You run several coding agents at once and want one place to watch and approve them
  • You want to review diffs and commit from the same view as the agent
  • You prefer to host your own tooling rather than use a hosted service