Meta launches Muse personal AI agent with a massive free usage tier, offering up to 100 million tokens per week, plus $20 and $100 paid plans for heavy users.

Meta launched a personal AI agent with a massive free usage tier

Meta launched a personal AI agent with a massive free usage tier, offering up to 100 million tokens per week at no cost, alongside two paid subscription plans at $20 and $100 per month. The agent, called Muse, was announced on September 8, 2026, and is initially available only in the United States for users aged 18 and over.

Muse is designed to act as a 24/7 digital assistant that can automate everyday tasks such as online shopping, booking movie tickets, scheduling appointments, filling out forms, and managing messages. It runs on Meta’s Muse Spark 1.3 model and is accessible via a dedicated app, the web at muse.ai, iOS, Android, and inside WhatsApp.

Free tier and paid plans

Meta is positioning Muse on a freemium ladder:

• Free tier: Up to 100 million tokens per week per user, as stated by Mark Zuckerberg. This is significantly higher than most competitors’ disclosed free allowances.
• Power plan: Reported at $20 per month, for users who need more compute and task automation.
• Maximum plan: Reported at $100 per month, for heavy users and advanced agentic workflows.

One unusual detail: multiple reports say Muse requires a payment card on file even to use the free tier, although users are not charged unless they exceed the free allowance or subscribe to a paid plan.

With one token roughly equivalent to four characters, 100 million tokens per week translates into a very large amount of text, commands, and agent interactions before hitting any paywall.

What Muse can do

Muse is built as a personal life‑management agent, not just a chatbot. It can:

• Book travel, restaurants, and events.
• Send emails and messages on your behalf.
• Manage calendars and schedule appointments.
• Fill out forms such as school permission slips.
• Coordinate multi‑step tasks across apps and services.
• Eventually interact with Meta’s AI glasses for voice‑first control.

The agent is intended to run continuously in the background, proactively handling routine tasks and only escalating to the user when necessary.

Security and confidential virtual machines

To address privacy and security concerns, Meta says Muse runs inside a confidential virtual machine (VM) architecture developed with input from Moxie Marlinspike, founder of Signal. According to reports:

• Each user’s agent runs in an isolated VM in Meta’s cloud.
• The VM is designed so that even Meta cannot directly inspect its contents.
• The architecture is marketed as “secure by design” for sensitive actions like payments and personal data handling.

This approach is meant to reassure users that their emails, messages, financial details, and other private information are not broadly accessible within Meta’s infrastructure.

Business model and long‑term vision

Zuckerberg has described Muse’s long‑term monetisation as a combination of:

• Subscription revenue from Power and Maximum tiers.
• A very small commission on transactions that Muse completes on behalf of users (for example, bookings, purchases, or reservations).

The generous free tier appears aimed at rapidly acquiring users and establishing Muse as a default personal agent before competitors such as OpenAI’s Atlas, Google’s Gemini Agent, and others gain similar traction.

Availability and next steps

At launch:

• Region: United States only.
• Age: 18+.
• Platforms: iOS, Android, web (muse.ai), and WhatsApp.
• Future: Integration with Meta AI glasses is planned but not yet dated.

Meta has not announced when Muse will expand to other countries or when additional languages beyond English will be supported.

Summary: Meta launched a personal AI agent with a massive free usage tier, offering up to 100 million tokens per week, plus $20 and $100 monthly plans. Muse runs on Muse Spark 1.3, uses confidential VMs for security, and is initially available in the US via app, web, and WhatsApp.