Meta (Muse Spark)
Add a Meta API key — from dev.meta.ai or from the credential Muse Code stored — and use Muse Spark through Memtro.
Meta's Muse Spark models reach Memtro through the Meta Model API, which speaks the OpenAI chat format. Add the key once and they work everywhere Memtro calls a model: chat, agent mode, routing ladders, memory extraction and jobs.
Add the key
On Dashboard → Providers pick Meta and paste a key. There are two places to get one.
From your Meta developer dashboard
Sign in at dev.meta.ai and create an API key. This is the key Meta bills for API usage, per token.
From Muse Code, if you already use it
Muse Code — Meta's coding CLI — stores the credential it signed in with, and the api_key inside it is an ordinary Meta API key. On macOS it lives in the keychain:
security find-generic-password \
-s "ai.meta.dev.credentials" \
-a "meta" \
-w
On Linux the same credential is the contents of ~/.config/muse/auth.json. Either way it prints one JSON object:
{"secret_schema_version":1,"api_key":"…","access_token":"…"}
Paste the whole object into the Meta field and Memtro takes the api_key out of it. The access_token beside it is a sign-in for the CLI itself, not a key the API accepts: a paste with only that in it is refused, and the form says so.
Model ids
| Id | What it is |
|---|---|
meta/muse-spark-1.3 | The current generation: a 1M-token context window, function calling and strict JSON schema output. |
meta/muse-spark-1.2 | The previous generation, and the cheaper rung of Memtro's default Meta ladder. |
auto/meta | Routed by difficulty within Meta — see model routing. |
Memtro's model list shows whatever your key can reach, which includes the -contributor variants. Those allow Meta to use the request to improve their products, so they are not among the ids Memtro suggests: a question asked through Memtro usually carries your organisation's memory with it.
What this is not
This is the API, billed per token. It is not the Muse Code subscription: that allowance belongs to the CLI on your own machine, and the access_token that unlocks it is not accepted by api.meta.ai. If you want a subscription rather than per-token billing, Claude Code and Codex are the two providers Memtro can run on one, because both ship a CLI Memtro can drive in a sandbox.