Meta Muse

Status: Available in the US since September 8, 2026, and in Japan since October 23, 2026. Some security and payment features are announced but not live yet (listed below).

A personal agent, not another chatbot

Meta already has a chatbot. Meta AI answers questions inside Facebook, Instagram, WhatsApp, and Messenger. Muse is a different product with a different job: you tell it something you want done, and it keeps working on it. Meta’s examples include sending emails, booking travel, filling out forms, lowering bills, planning a party and sending the invites, and turning a recipe reel into a grocery list.

Muse helps you turn a goal into a plan, then moves the work forward on its own, coming back when something changes or when it needs your approval. You talk to it in the Muse app or in WhatsApp, the same way you would message a person. You can rename it and change its avatar and communication style.

How the Muse product family fits together

Meta now uses “Muse” for several different things, which makes searches confusing. Here is how they relate as of October 2026:

Meta’s launch post says Muse is powered by Muse Spark. When Meta released the Muse Wonder models on October 12, it said Wonder would “soon” power more experiences in Muse, starting with subscribers, without giving a date. Until Meta confirms it, assume the agent you are using still runs on Spark.

Where Muse does its work

Each user’s Muse runs on a dedicated cloud virtual machine that Meta calls Muse Secure VM. The VM holds both the agent and your data, has its own browser, and is isolated so other users’ agents cannot reach it. That is the same basic idea as an agent sandbox: give the agent its own computer instead of yours.

The more unusual piece is Sentinel, a second agent that runs on the same machine but is separated from Muse at the system level. According to Meta, Muse cannot reach the internet unless Sentinel approves the action, and Sentinel asks you when it needs to. The design puts a gatekeeper outside the agent doing the work, which is a sound principle. How well it holds up is still an open question; TechCrunch noted at launch that Meta’s technical claims need deeper review by outside security researchers.

Connected apps and access levels

You connect services one at a time and decide how much access each gets. Email is the clearest example: you can let Muse only read mail, or also send on your behalf. Meta lists categories including email, calendars, payments, health and fitness, smart home, dining, shopping, music, and events. Where a service has a public API, you can connect it with your own credentials. Where it does not, Muse uses its browser.

You can change access or disconnect a service at any time, and you can tell Muse to forget specific things it has learned. Muse also keeps an audit trail of actions it has taken and actions it plans to take, which is the first place to look when something seems off.

Approvals and purchases

Muse asks for approval before sensitive actions such as sending an email or making a purchase. Engadget’s setup guide suggests switching approvals to “always ask” while you get used to it, which is reasonable advice given that Meta itself says Muse can make mistakes.

Checkout uses Link by Stripe. Link generates a one-time card for each purchase, so your real card number is not shared, and Meta says Muse is the first AI agent covered by Link’s purchase protections on eligible orders. Meta says Muse cannot see your passwords or payment methods, including passwords you type into its browser yourself.

Privacy and training

This is where most of the scrutiny has landed. Muse is designed to connect to email, calendars, payments, and health services, and it learns from your conversations to make suggestions you did not ask for. Meta makes three specific commitments:

1. You can opt out of having your Muse interactions used to train Meta’s AI models. Training appears to be on by default, so check it during setup.
2. Muse does not share your conversations or VM data with Meta’s ad systems.
3. Muse cannot see your passwords or payment details.

Opting out of training does not stop Meta staff from accessing data where needed to support, secure, or operate the service, according to third-party guides that walked through the settings. And conversations you have with Muse about other people, such as email threads, involve their data too. Read our AI data retention explainer before connecting accounts that hold client or family information.

Announced versus available

As of late October 2026, these are available: the core agent, iOS and Android apps, muse.ai, WhatsApp access, built-in connectors, Secure VM, Sentinel, and Link checkout.

These are announced but not yet available: Muse Confidential VM, which Meta says will encrypt the whole VM with a key only you hold so that even Meta cannot access it (“later this year”); Shop Pay checkout; AI glasses support; and 1Password sign-in. 1Password for Agents lists Muse as rolling out from October 22, so that one may reach your account soon.

Pricing

Muse is free to start with limited usage, and the app shows a usage meter. TechCrunch reported two paid tiers at launch, Power at $20 per month and Maximum at $100 per month, both adding usage. A payment card is required to sign up even for the free tier. Meta expects most people to stay free.

Who it suits

Muse is the most accessible of the new personal agents: it has a free tier, works inside WhatsApp, and is aimed at everyday errands rather than office work. That makes it a reasonable way to see what an ongoing agent feels like. It is a weaker fit if you are uncomfortable giving Meta deep access to email, payments, and health data, or if you need it outside the US and Japan.

Alternatives

• OpenAI Dots, always-on agents inside ChatGPT, on higher-priced plans
• Claude Duet, Anthropic’s proactive agent, in US research preview
• OpenClaw, a self-hosted agent where you control where the data lives
• Manus, a general task agent

For the concepts behind this category, see proactive agents, persistent agent state, authorization gates, and AI permission hygiene.

Compare: OpenAI Dots vs Meta Muse, Claude Duet vs Meta Muse, and the three-way Dots vs Duet vs Muse guide.