# Meta Muse **Aliases:** Muse, Meta Muse personal AI agent **Entity class:** Personal AI agent **Developer:** [[wiki/Meta|Meta]] **Introduced:** September 8, 2026 **Maturity:** Evolving ## Definition **Muse** is Meta's consumer [[wiki/Personal AI Agent|personal AI agent]]. It is designed to receive goals conversationally, work across connected applications and the web, perform multi-step tasks, continue in the background, remember user context, and return for approval before selected consequential actions. This node is specifically about Meta's 2026 personal agent. It is distinct from [[wiki/MUSE|Promise's MUSE studio system]], [[wiki/Muse EEG|Muse EEG]], and [[wiki/Poet's Muse|the literary concept of the muse]]. ## Architecture Meta describes each Muse as running in **Muse Secure VM**, a dedicated virtual machine with its own browser. The product connects to applications and services, stores credentials behind a boundary the agent cannot directly read, keeps task state after the app closes, and presents an audit trail of completed and planned activity. A separate Sentinel agent evaluates outbound activity and requests permission when needed. Muse is powered by Meta's Muse Spark model family. Its consumer stack combines several layers that are often separate in developer systems: - planning and multi-step tool use; - browser and connected-application action; - persistent runtime and [[wiki/Agentic Memory|memory]]; - credential isolation; - [[wiki/Approval Architecture|approval]] before actions such as sends and purchases; - payment through one-time card mechanisms and supported wallets; - mobile, web, and WhatsApp interfaces. ## Product Lineage [[wiki/Facebook M|Facebook M]] tested the same intent-to-outcome product abstraction from 2015 to 2018 but depended heavily on human trainers. Muse represents Meta's return to the personal-agent category after planning models, browser competence, memory, virtualized runtimes, and payment and approval infrastructure changed the supply side. [[wiki/OpenClaw|OpenClaw]] is an important adjacent lineage and influence on the category: it made persistent, tool-using personal agency visible as open infrastructure for technical operators. Muse changes the distribution model by packaging comparable primitives behind consumer onboarding and a managed secure runtime. [[wiki/Grok Bot|Grok Bot]] is a closer hosted competitor organized around a roster of persistent named agents and routines. ## Human Execution Boundary Reuters reported in September 2026 that Meta had internally tested trained human contractors placing some business calls requested through Muse. Employees raised disclosure and privacy concerns, and Meta rolled back the test. Reported internal figures placed human completion at 95–98 percent, above an unspecified AI-only rate. The episode repeats one element of Facebook M at a narrower boundary. It does not mean that Muse as a whole is human-operated. It shows that agentic systems can remain machine-led while concentrating human labor at interfaces where institutions resist synthetic callers. It also makes operator disclosure part of [[wiki/Approval Architecture|approval architecture]]: permission to delegate to software is not automatically permission to disclose the task to a contractor. ## Competitive Position - **[[wiki/OpenClaw|OpenClaw]]:** open-source, self-hostable, model-pluggable, operator-controlled infrastructure; more configurable and technically demanding. - **[[wiki/Grok Bot|Grok Bot]]:** hosted persistent bots organized as a team roster with computers, role memory, tools, skills, and routines. - **Muse:** managed consumer agent centered on one person's life, connected applications, secure virtual-machine isolation, approvals, and mass-market distribution. These systems overlap in delegated browser action, memory, persistence, and tool use, but they compete through different trust, ownership, and interface models. ## Relationships - **developer:** [[wiki/Meta|Meta]]. - **historical precursor:** [[wiki/Facebook M|Facebook M]]. - **category:** [[wiki/Personal AI Agent|Personal AI Agent]], [[wiki/Agentic AI|Agentic AI]], and [[wiki/AI Counterpart|AI Counterpart]]. - **operational stack:** [[wiki/Agentic Memory|Agentic Memory]] and [[wiki/Approval Architecture|Approval Architecture]]. - **comparators and competitors:** [[wiki/OpenClaw|OpenClaw]] and [[wiki/Grok Bot|Grok Bot]]. - **human consequence:** [[wiki/Agentic Empowerment|Agentic Empowerment]]. ## Sources / Provenance - [Meta — Introducing Muse: The World’s First Personal AI Agent Built for Everyone](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) - [Meta AI — Muse product and architecture overview](https://ai.meta.com/muse/) - [Meta — What is agentic AI?](https://ai.meta.com/learn/agentic-ai/what-is-agentic-ai/) - [Reuters — Meta testing a human concierge for Muse, September 22, 2026](https://finance.yahoo.com/news/exclusive-meta-testing-human-concierge-160132946.html) - [[wired/Wired About the Age of Agentic Empowerment|Wired About the Age of Agentic Empowerment]] - [[wired/Wired About Personal Agents Named M|Wired About Personal Agents Named M]] **As of:** 2026-09-28