What is Manus, the agent Meta tried to buy for $2B?

Manus is an AI agent that takes on a task and delivers a finished result — a website, a slide deck, working code, a research report — instead of just answering in a chat window. It's also the AI startup Meta tried to buy for roughly $2 billion in December 2025, only for Chinese regulators to block the deal on national-security grounds and force Meta to walk away in June 2026. Manus now runs independently out of Singapore under its original founders. Expectations across the industry are being reset, though: Zuckerberg has told Meta staff that agent progress is slower than hoped.

Definition: Manus is an autonomous AI agent that plans and executes multi-step tasks — research, coding, design, browser actions — and hands back a finished deliverable.

Example: Give Manus a prompt like "build me a one-page website for a coffee shop" and it plans the steps, writes the code, and returns a working site, rather than just describing how to build one.

Key takeaway: Manus is positioned as execution-first, not conversation-first — the distinction it draws against general chatbots.

Business impact: Teams considering Manus should evaluate it like any third-party agent platform: real capability, but a credit-based cost model and a data/permissions setup worth checking before connecting real accounts.

What does Manus actually do?

Manus positions itself as a general-purpose autonomous agent rather than a chatbot: you describe an outcome, and it plans and executes the steps needed to produce it. Its feature set spans generating slide decks and websites, assisting with coding and desktop app builds, producing images and design assets, running a browser autonomously to complete web-based tasks, conducting multi-source research through a feature it calls "Wide Research," and connecting to tools like email and Slack to act on real accounts rather than only inside its own interface.

The product's own framing — "less structure, more intelligence" — captures the pitch: instead of a rigid workflow builder, Manus is meant to figure out the steps itself from a loosely specified goal.

Who actually built Manus?

Manus comes from Butterfly Effect Pte. Ltd., founded by Xiao Hong in 2022. Co-founder Ji Yichao ("Peak") serves as chief scientist. The product launched in invitation-only beta in March 2025, with a demo showing it autonomously completing tasks like resume screening and stock analysis that reportedly drew over a million views within 20 hours. The name itself references MIT's motto "Mens et Manus" — mind and hand.

The company raised a seed round in 2023 (ZhenFund), a Series A in November 2024 (Sequoia China, Tencent), and a roughly $75 million Series B in April 2025 led by Benchmark, reaching an estimated $500 million valuation at the time. Around the same period, the company relocated its headquarters from Beijing and Wuhan to Singapore.

What happened with the Meta acquisition?

Meta announced its acquisition of Manus on December 30, 2025, in a deal reportedly valued between $2-3 billion. China's Ministry of Commerce opened a review in January 2026 examining export controls, technology transfer, and foreign investment rules. Scrutiny escalated through March, and on April 27, 2026, China's National Development and Reform Commission formally blocked the deal and ordered it unwound. Meta announced it was officially cutting ties on June 15, 2026, citing the regulatory obstruction.

Manus continues to operate as an independent company from Singapore, under its original founders, reportedly at a run rate around $125 million in annual revenue by the end of 2025. The episode is a concrete example of how cross-border AI acquisitions are now subject to national-security-level review on both sides, not just ordinary antitrust process.

How is Manus priced?

Manus uses credit-based pricing rather than a flat per-seat fee: every action the agent takes — browsing a page, running code, generating a file — consumes credits, so a simple lookup costs far less than a multi-step research task with browser navigation and a formatted report. As of mid-2026, commonly reported tiers include a free plan with a limited daily credit refresh, and paid plans in the range of roughly $20, $40 and $200 per month with progressively larger credit allowances, alongside custom Team and Enterprise pricing with seat-based billing and admin controls. Credit-based pricing shifts more often than flat subscriptions, so treat exact figures as directional and confirm current pricing on Manus's own site before budgeting around it.

How does Manus compare to a general chatbot?

This is the same underlying distinction covered in AI agents vs. RPA: Manus sits closer to the flexible, judgment-driven end of that spectrum than to fixed, rule-based automation.

Should a business build this in-house or use Manus?

That's a real build-vs-buy decision, not a given. Manus offers broad, ready-made execution across many task types without engineering investment, which suits teams that want to try agentic workflows quickly. Teams with specific, repeatable workflows and existing engineering capacity may still get better cost and control building a narrower, purpose-built agent instead of paying per-credit for a general-purpose tool. Neither answer is universally correct — it depends on how repeatable the task is and how much control over data and process actually matters for that specific workflow.