Does Muse AI Have an API? Meta Model API vs the Muse Agent
Muse the agent has no public API — but Meta's Model API does expose Muse Spark models. Here's the difference, what developers can build today, and what's still closed.
Searches for a "Muse AI API" usually mean one of two different things, and the answer is opposite for each: the models are available to developers; the agent is not. This page separates them, because conflating them is the fastest way to buy the wrong thing or build on a false premise. For the vocabulary itself, see our guide to Muse's new terms.
The short answer
- Muse the agent — no API. Meta has announced no way to programmatically drive Muse. Nobody outside Meta can make Muse send email, book travel, or fill out forms via code. The agent is reachable only through the app, muse.ai, and WhatsApp.
- Muse Spark models — yes, via the Meta Model API. Meta operates a developer API with Muse Spark models behind it, and the models are additionally reachable through OpenRouter.
- Muse Glimmer — downloadable outright. An open-weights, 30B model under Apache 2.0 for local agents.
The Meta Model API: what developers get
Meta's developer hub (developer.meta.com, with docs at dev.meta.ai) exposes the Muse Spark family — the same model lineage that powers the agent. Meta's announcement of Muse Spark 1.1 describes a multimodal reasoning model built for agentic tasks, with gains in tool use, computer use and coding; Meta AI chief Alexandr Wang described it as "trained to know when to automate and when to use interfaces directly — it scripts when that's faster, clicks when direct." The model catalog also includes the open Muse Glimmer (30B, Apache 2.0, single-GPU) and Muse Voice Transcribe for real-time speech recognition.
What you get with the Model API is model intelligence: reasoning, tool calling, computer-use behaviors, transcription. What you do not get is everything that makes Muse feel like a product:
- No Secure VM — the isolated cloud computer where the agent runs
- No Sentinel — the guardrail agent that gates everything reaching the internet
- No approval flows or audit trail — the human-in-the-loop layer
- No service integrations — email, calendar, payments (Link by Stripe), shopping
That product layer is precisely Muse's differentiation, and Meta has kept it closed. If you want to prototype agent behavior, the API plus your own orchestration is a legitimate path — but you are building the guardrails yourself, which is a serious undertaking Meta needed a dedicated second AI for.
What's still closed (and what to watch)
Everything agent-level: task execution on connected accounts, the VM, approvals, integrations, and Muse's memory system. There has been no announcement of an agent-tier API, waitlist, or enterprise program. Watch this page and our updates changelog — an agent API, if it ever ships, would most likely follow the same path as the Model API: developer announcement first.
If your actual goal is not development but getting more out of Muse as a user, the lever isn't an API — it's better task briefs. Our prompt generator builds structured instructions you can paste into Muse (it doesn't connect to Muse directly; nothing outside Meta's apps does).
Keep reading
Frequently Asked Questions
Does Muse AI have a public API?
No. There is no public API that operates Muse the agent — no way to programmatically tell Muse to send your email, book a trip or run a task. The agent is only reachable through its official surfaces: the Muse app, muse.ai, and WhatsApp. What does exist is Meta's separate Model API for Muse Spark models.
What is the Meta Model API?
It's Meta's developer API (at developer.meta.com, docs at dev.meta.ai) that provides programmatic access to Muse Spark models — the model family that powers Muse — with multimodal input and tool-use capabilities. Meta has also listed the models on OpenRouter. It's an API to the models, not to the agent built on top of them.
Can I build my own agent like Muse with the API?
Partially. You get the model — tool use, computer-use capabilities, multimodal reasoning — but none of Muse's product layer: no Secure VM, no Sentinel guardrail, no approval flows, no service integrations, no audit trail. You'd be building that infrastructure yourself.
Can I run any Muse model locally instead of using an API?
Yes — Muse Glimmer, a 30-billion-parameter model released under the Apache 2.0 license, is downloadable from Hugging Face and LM Studio and runs on a single GPU. The main Muse Spark models are cloud-only through Meta's services.
Official sources
This is an independent guide and is not affiliated with or endorsed by Meta.
Put Muse to work
Browse ready-made task guides or generate a copy-ready Muse instruction in seconds.