Connect Muse Code
3 minute read · Chats & Local Agents
Use Muse Code from Meta as a coding agent in a Vortex conversation. Your Meta account supplies the agent and model access; Vortex keeps the project, tools, approvals and review beside the task.
- Connect an account
- Choose the access mode
- What to expect from Muse Code
- Read its usage
- Check the result
Connect an account
- Open Settings → AI providers, add a provider and choose Meta · Muse Code under Choose a provider.
- If Muse Code is not installed, Vortex shows Installing Muse Code…. It downloads Meta's native build from the stable channel, checks its size and checksum, and verifies the installed version. Use Cancel installation to stop.
- Sign-in starts when the installation is ready. Vortex opens Meta's sign-in page and shows a code. Enter that code on the page with Copy code if useful, and finish in the browser.
- Check that the account row shows it is connected. In a conversation, choose Muse Code, the intended account and an available model.
- Start with a small read-only request: “Inspect this project and explain how to run it.”
Builds are offered for macOS, Windows and Linux on x64 and arm64 processors. Linux needs a glibc-based distribution. Updates come from the same channel.
You can add several Muse Code accounts. Each one keeps its own sign-in, separate from the others and from a Muse Code login elsewhere on the computer. Use Sign in again on the account row when its sign-in expires. To hide Muse Code from the model menu for new chats, turn off Show Muse Code in the model menu under Model menu; existing chats keep their provider.
Choose the access mode
Muse Code applies its own permission profile for each access mode. Vortex checks that the profile was applied; otherwise the message is not sent.
| Access mode | Muse Code behaviour |
|---|---|
| Supervised | Read-only: it cannot write files or run shell commands. Choose another mode when you want changes. |
| Plan | Read-only. |
| Auto-accept edits | Asks before each action that needs approval. |
| Auto | Uses Muse Code's own automatic review. |
| Full access | Unrestricted, with its sandbox turned off. Use it only for work you have deliberately authorized. |
An approval card titled Muse Code: with the tool name shows the command or path. Each approval covers that single action. Answering a Muse Code has a question card supplies information; it does not approve unrelated work. See permissions and approvals.
What to expect from Muse Code
- Plan with
/planand/default, model choice and reasoning effort where the model supports it. - Images as message attachments.
- Follow-up text can steer the active turn. A follow-up that contains an image waits for the next turn instead.
- Vortex tools such as Preview, saved connections and linked tasks, when the installed version supports them. If it does not, Vortex shows Update Muse Code in AI connections.
- Local scheduled tasks.
Muse Code does not offer the provider's own slash commands in Vortex, /compact or /review, or SSH-hosted runs.
Read its usage
The usage indicator can show a five-hour window and a weekly window with reset times, and a quota error when the account is exhausted. Until Muse Code reports values, Vortex shows Muse Code has not reported account limits yet. Missing information does not mean unlimited usage. See models and provider usage.
Check the result
Ask for one visible change, inspect its diff in Files and Source Control and try the app in Preview. A finished message does not prove the app works. If sign-in fails, start Sign in again for the same account. If a run stops, read its last completed step and the changed files before retrying.