How YOLO Studio runs Muse Code
YOLO Mode, with the workspace as the sandbox
Muse Code launches with --yolo, so it edits, runs and tests without asking before each step. Its own shell sandbox cannot run inside a workspace virtual machine, so it is switched off: the workspace already isolates this one project.
Installed the first time you open it
Muse Code is not built into the workspace image. The first time you add a Muse Code tile, the workspace installs it with Meta’s installer, and from then on it starts like any other agent.
Workspace tools over MCP
Muse Code comes connected to YOLO Studio’s MCP server. It can open preview tiles, create and update Kanban cards, ship to a yolo.host address, and read or message other agents’ tiles, with a token scoped to that one session.
In the Background Runner too
Pick Muse Code as the agent for background work and the Background Runner hands it cards from your Kanban board, each in its own lane, alongside your other agents.
What the workspace gives it
A sandbox, not your laptop
Every workspace is a secure Linux sandbox running in its own virtual machine. The agent works on your repository inside it, as a non-root user, with your files kept between sessions. It cannot reach your laptop, your other projects or anyone else’s workspace.
One lane per agent
Start an agent in its own lane and it gets its own git clone on a lane/<name> branch, so agents working side by side never share a git index or step on each other’s changes. The Quick Lane button starts your primary agent in a fresh lane in one click.
See the result running
Run your app in a preview tile next to the agent. When the agent saves a file, the preview reloads, and one command ships what is in the preview to a yolo.host address.
Git that just works
Connect a GitHub repository (or host one inside YOLO) when you create the workspace. It is cloned and configured for you, so git commit, git push and gh pr create work in the agent’s session without extra setup.
Getting Muse Code running
Create a workspace
Sign in with GitHub, choose New Workspace, and pick the repository to work on.
Add a Muse Code tile
Choose Muse Code from the tile picker. The first time, the workspace installs it for you; after that it starts like any other agent. Pick the main workspace or a new lane to run it in.
Connect your access
Add your Meta API key from Agent access. Credentials are delivered to the session, never shared with other workspaces or users.
Give it the job
Type into the tile, or drag files onto it. The shortcut bar sends common follow-ups in one click: Commit & Push, Pull Latest, Merge to Main and Continue.
For more than one task at a time, put the work on a Kanban board and let the Background Runner hand cards to Muse Code and your other agents, each in its own lane. The guides below walk through it.
Questions
What do I need to use Muse Code?
A Meta API key, added once from Agent access. Muse Code signs in with that key; there is no separate browser sign-in to complete inside the workspace.
Is Muse Code pre-installed?
No. It installs the first time you open a Muse Code tile. Claude Code, Codex CLI, Gemini CLI and OpenCode are the agents built into the image.
Can I run it alongside other agents?
Yes. Claude Code, Codex CLI, Gemini CLI and OpenCode are pre-installed, and agents such as Muse Code, GitHub Copilot, Aider, Goose and Qwen Code install the first time you pick them. Put each one in its own lane and they work on the same repository without colliding.
Does it stop when I close the tab?
No. The agent runs in the cloud workspace, not in your browser, and closing its tile does not stop it. Cards queued on a Kanban board keep moving through the Background Runner. A workspace left idle long enough is paused, with its files kept for when you resume.
Keep reading
Try it on your own repository
Sign in with GitHub, open a workspace on your repository, and put your first agent to work in a few minutes.
Get started