The Phaseo CLI can launch Muse Code with a dedicated gateway key and temporary provider settings.
Set up with the Phaseo CLI
Use --catalog all to expose every active Phaseo text/chat model, or --dry-run --json to preview the launch plan without creating a key or starting Muse Code:
Use phaseo run muse ... for the explicit command form. Put Muse Code flags after -- so Phaseo can consume its own flags first.
On Unix-like systems, the first interactive run offers to install Muse Code with Meta’s installer. On Windows, install Muse Code separately, verify muse --version, and use --skip-install.
Manual configuration
Muse Code’s public setup is built around Meta’s own Model API and authentication flow. Meta does not currently document a stable persistent setting for replacing that endpoint with a third-party OpenAI-compatible gateway, so the Phaseo CLI runner is the supported way to route Muse Code through Phaseo.
If you want to use a Phaseo model from another client manually, use the OpenAI-compatible integration with https://api.phaseo.app/v1 and a Phaseo API key. Do not copy the runner’s temporary Muse settings into your normal Muse profile; those files are implementation details and are removed when the runner exits.
Verify
Send a small prompt through the launched harness, then confirm the request appears in Phaseo activity. The runner writes settings.json under a temporary XDG_CONFIG_HOME and removes it when Muse Code exits.
Troubleshooting
Muse Code is not installed
Install Muse Code separately, verify muse --version, and rerun with --skip-install.
The model is not listed
Use --catalog all or pass a model id with --model. Model discovery is not required for the Phaseo gateway.
Last modified on September 10, 2026
