Meta launched Muse on 8 Sep 2026 — a personal AI agent that books travel, sends emails, fills forms, and processes payments on the user’s behalf, available in the United States on iOS, Android, and muse.ai at $20 or $100 per month.
What Meta Muse Does
Meta Muse executes multi-step tasks with minimal supervision. The agent connects to a user’s email, calendar, and payment methods, then carries out work such as drafting and sending emails, booking travel, submitting forms, setting appointments, and negotiating on the user’s behalf. Muse continues working after the user closes the app and returns only when a consequential action — such as sending a message or completing a purchase — requires explicit approval.
Muse is also accessible via WhatsApp. Users can send Muse a task by messaging it the same way they would message another person, giving the agent access to WhatsApp’s more than 2 billion existing users without requiring a separate app download.
How Muse Works: The Secure VM Architecture
Muse runs on the Muse Secure VM, a dedicated virtual machine that houses the agent and the user’s data in isolation rather than in a shared cloud fabric. Meta positions this architecture as a direct response to the privacy concern that has slowed enterprise AI agent adoption — user data does not co-mingle with other workloads.
The underlying model is Meta Muse Spark, the same foundation model the company has made available via API for developers. Muse Spark is currently Meta’s most affordable API-accessible model at $0.30 per million tokens combined.
Muse Pricing: Free, $20, and $100 Per Month
Meta offers Muse at three pricing tiers: a free tier, a standard plan at $20 per month, and a pro plan at $100 per month. Payments within the app are handled through Stripe Link, which Muse integrates with directly. Meta has not disclosed which specific features are gated to each paid tier beyond the “standard” and “pro” labels.
At $100 per month, the pro plan places Muse in the same pricing bracket as Salesforce Starter and comparable business software — making it a direct alternative to virtual assistant services for solopreneurs and small business owners managing repetitive operational tasks. For a broader look at how Muse fits among today’s best AI agents for business tasks, including how autonomous agents compare across use cases, the site’s dedicated hub covers the field in depth.
Meta’s Distribution Bet: WhatsApp and AI Glasses
Meta has announced plans to add Muse to Meta AI glasses, though the integration is not yet live. If delivered, Muse would become the first widely distributed AI agent accessible without a phone screen — a distribution channel OpenAI and Anthropic currently lack.
Meta’s stated positioning is mass-market accessibility rather than enterprise-first adoption. According to Meta’s official blog post, cited by PYMNTS: “Unlike other agents, Muse was built to work for billions of people worldwide, so there’s no learning curve. Anyone can use it out of the box, no technical experience required.” The company’s broader framing, stated by Meta leadership since at least April 2026, is that personal superintelligence will be among the most transformative technologies of a generation — and Muse is described as a first step toward that goal.
Muse competes directly with OpenAI’s operator-mode and Anthropic’s Claude computer-use capabilities, both of which are aimed at similar task-automation use cases but have so far prioritized developer and enterprise access over consumer distribution. For concrete examples of how AI agents are already automating business workflows, see the site’s overview of real-world AI agent examples across industries.
Our Take
Meta’s distribution advantage is real: WhatsApp reach at launch plus a roadmap to AI glasses gives Muse a consumer acquisition path that neither OpenAI nor Anthropic has matched. The Secure VM architecture addresses the single biggest objection enterprise buyers have raised about AI agents acting on their behalf. The outstanding question for business users is whether asynchronous, approval-gated operation is sufficient oversight — or whether the agent needs tighter guardrails before it handles anything beyond low-stakes tasks.
For Context — Meta Muse Model Coverage: This article covers the Muse consumer app launch. For coverage of the underlying Muse Spark model and its API pricing, see Meta Muse Spark 1.1 Is Here — and It’s Cheaper Than Claude or GPT. For the Muse coding agent, see Meta Muse Code Launches: AI Coding Agent 21x Cheaper to Try. For the open-weight release, see Meta Muse Glimmer: Open 30B AI Agent That Runs Without Cloud.
