muster
Muster Agent · desktop app
The desktop app that works with Codex, Claude Code, OpenCode, the Cursor Agent CLI, the Gemini CLI, any OpenAI-compatible router and local models, with long-term memory, sandboxed computers, live diffs, a terminal and a browser.
v0.3.11 · free · zip, tar.gz & SHA256SUMS
Install
Open the .dmg and drag Muster Agent onto Applications. The first time, right-click it and choose Open (or System Settings › Privacy & Security › Open Anyway). Updates then install themselves when you restart. Needs macOS 14+ on Apple silicon.
Run the -setup.exe. It installs for your user, with no admin prompt, and adds a Start menu shortcut. If SmartScreen warns, choose More info › Run anyway. Terminals open PowerShell. A portable .zip is also on the release.
chmod +x Muster-Agent-*.AppImage and run it on any x64 distro. On Debian or Ubuntu: sudo apt install ./Muster-Agent-*-linux-amd64.deb. Sandboxes use Docker Engine.
Prefer the terminal?
The open source Muster agent harness has a terminal chat with slash commands, @agent routing, memory controls and MCP setup. The demo needs no model key.
pnpm --package=@musterhq/cli dlx muster demo
Works with
Found automatically when installed on your computer, added as a provider, or read from a gateway named in your Codex config. Each person uses their own key. Read the guide.
OpenRouter has its own provider card. LiteLLM and any OpenAI-compatible gateway or custom endpoint, such as vLLM or a company proxy, work through the OpenAI-compatible connection.
Ollama and LM Studio are detected on their default ports. Muster lists whatever models your router or provider offers, so new models appear automatically.
It also runs Codex (your ChatGPT sign-in), Claude Code, OpenCode, the Cursor Agent CLI and the Gemini CLI that you have installed and signed in to. Any OpenAI-compatible API works too, including Meta’s Muse models (Muse Spark, Muse Wonder) through the Meta Model API, plus open-weight models you run locally.
Models you can use
OpenAI GPT via Codex or the API, Anthropic Claude via Claude Code or the API, Google Gemini via the Gemini CLI or the API.
Meta Muse via the Meta Model API, xAI Grok via the Grok CLI.
DeepSeek, Qwen, Mistral, Kimi, GLM and Llama or other open-weight models, through a router or run locally.
Teams and containment
Check a task out of a Paperclip org, work locally in a git worktree, and let the work hand itself back after a 60 second countdown. Read the guide.
Give a chat a disposable Docker computer with no network by default and capped memory, CPU and processes. Docker is optional. Read the guide.
Read-only, workspace or full access per chat, with secrets kept out of the commands the agent runs.
FAQ
Yes. Muster Agent is free and open source under the MIT licence. You bring your own model access: ChatGPT or Claude sign-ins, API keys, OpenAI-compatible gateways, or local models through Ollama and LM Studio. It finds the ones already on your computer, so you do not sign in again.
Codex (your ChatGPT sign-in), Claude Code, OpenCode, the Cursor Agent CLI and the Gemini CLI, plus any OpenAI-compatible router (OmniRoute, OpenRouter, custom endpoints) and local models through Ollama and LM Studio. It runs the installs you already have side by side and lets you pick the model per chat.
macOS 14 or later (Apple silicon and Intel .dmg), Windows 10 and 11 x64 (installer .exe or portable .zip), and x64 Linux (.AppImage, .deb or .tar.gz).
Muster Agent runs on your computer and works in your own folders. Your code goes only to the model provider you choose for a chat, so a local model keeps everything on your machine. Optional Docker sandboxes have no network by default.
Muster Agent is an independent open-source project, not made by or affiliated with OpenAI or Anthropic. It is a desktop workbench that works with the Codex and Claude Code installs and sign-ins you already have.
On macOS it downloads the new release, checks its SHA-256 and code signature, and installs on restart. On Windows and Linux it tells you when a new version is out and opens the download.