How does Meta Muse's memory work, and can it use outside memory?

According to Meta's Help Center, Muse builds its understanding of you from four sources:

• information you share directly in conversations;
• preferences and patterns it observes from your interactions;
• context from connectors you've connected, such as emails and calendar events;
• its file-based memory system.

Meta's launch post describes the goal as remembering “what matters to a person, so it can make suggestions unprompted and act on details that person only mentioned once.”

1. 1Step 1: See it: ask Muse what it remembers, or tap the Assistant icon, then Identity, then Memory.
2. 2Step 2: Edit it: ask Muse to change something, or edit the information manually. You can ask Muse to revert a change.
3. 3Step 3: Forget: ask Muse to forget a topic or a set of interactions.
4. 4Step 4: Start over: Settings → Data Controls → Reset Muse.

Meta's Help Center notes that after you delete something, Muse may still remember information it learned from it. The setting that lets your interactions help develop Meta's AI models is on when you first use Muse; you can change it.

Meta says Muse checks with you before sensitive actions like sending an email or making a purchase. For connector permissions, third-party guides describe scoping access to one action, one task, a time window, or permanently.

• Directory connectors are reviewed by Meta and listed in Settings. Meta opened submissions to developers on September 18, 2026, through its connector platform.
• Custom connectors are ones Muse sets up for you when you ask in chat. Meta's help center says it doesn't review custom connectors or how they use your information, so grant access with caution.
• Muse adds a custom connector through a Secure credentials store sheet with two fields: the connector's address and a Client ID, the identifier an OAuth sign-in uses. Some third-party guides published in September 2026 describe the flow as built around API keys; the form itself shows OAuth is supported.
• The same sheet says: “Info from this connector is part of your AI interactions, which we use to improve AI at Meta.” Anything Muse reads through a connector can be used that way.

Not confirmed yet; it is being tested now. Morphika has no API keys by design: every agent signs in through the browser with OAuth 2.1 and PKCE, gets a short-lived token, and can be revoked on its own. Muse's custom connector form uses OAuth, and Morphika supports dynamic client registration, so the pieces fit. This page will carry the exact steps once they are confirmed.

If you try it: use a read-only connection first (https://mcp.morphika.ai/read/mcp). Muse reads email and the web, and Meta says connector information may be used to improve its AI, so start with memory Muse can read but not change.

Meanwhile, the same memory already works across Claude Code, Codex and Claude Desktop: see one memory for all your agents.

• Prefer OAuth sign-in over pasting a long-lived API key.
• Prefer a read-only connection if Muse only needs to look things up.
• Check whether the service keeps an audit log of what the agent read and wrote.
• Check how to revoke the connection, and that revoking takes effect at once.
• Remember that anything a connector returns can become part of Muse's own memory.

The full checklist is in how to keep an AI agent's memory secure.

• Meta: Introducing Muse (September 8, 2026)
• Meta Help Center: Customize Muse's personality and memories
• Meta Help Center: Manage your Muse data
• TechCrunch: Everything new coming to Meta's AI agent Muse (September 23, 2026)
• CellCog: Muse connector platform (September 19, 2026)
• Parallel: Meta Muse custom integrations
• Parallel: Best MCP servers and connectors for Meta Muse

Sources checked September 25, 2026.

• Instinct AI and memory: what it keeps and how to stay safe
• How to give Claude Code persistent, secure memory
• How to give OpenAI Codex persistent, secure memory