Manus is a general-purpose AI agent from Butterfly Effect, a Singapore company, that takes a task from a user and runs it on its own cloud computer, browsing, researching, filling forms and building documents or apps along the way. It opened as an invite-only preview on Mar 6, 2025 and to public registration on May 13, 2025, and it has since added a Browser Operator extension (Nov 18, 2025) that lets the agent work inside the user's own browser, plus a desktop app.
The company has had an unusual year. Meta announced a deal to buy Manus in Dec 2025, China ordered it withdrawn on Apr 27, 2026, and Manus has run independently since Sep 1, 2026. On Sep 28, 2026 it launched Manus 2.0 and the Cue app, which give agents their own email addresses, phone numbers, wallets and computers, and on Oct 8, 2026 it raised over $500M led by Boyu Capital and IDG Capital. One public real user monitoring dataset on GitHub counted 9,000 to 20,000 Manus requests a day across 224 to 341 sites in the week of Oct 24, 2026, which its authors call a lower bound.
Site owners ask about Manus for two reasons. Its users send it to do real work like research and bookings, and the same agent is good at scraping and at creating accounts, which the new Cue identities make easier.

How Manus shows up on your site
In its default cloud mode, Manus browses from an isolated virtual computer. Researchers at UC Davis found about 96.6% of the Manus tasks they studied ran Linux Chromium with the clock set to UTC, and the GitHub study traced the traffic to Google Cloud. The user agent is a normal Linux Chrome string with Manus-User/1.0 appended, and page loads carry a Web Bot Auth signature naming https://api.manus.im, with the Ed25519 public key published in Manus's own key directory. Cloudflare Radar lists the agent in its bot directory as Manus Bot and points to that same key directory, which is a useful cross-check that the signature belongs to Manus.
| Check | What Manus does |
|---|---|
| User agent | Linux Chrome with Manus-User/1.0 appended |
| Web Bot Auth signature | Signs page loads as api.manus.im |
| IP ranges | No published list, seen on Google Cloud |
| robots.txt | Not documented by Manus |
Browser Operator and the desktop app behave differently, because they drive the user's own browser on the user's own network and carry no Manus token at all.
Why Manus is only partly easy to detect
Cloud Manus is one of the more honest agents on the web, since it declares itself in the user agent and signs its requests. The catch is in the details. The GitHub study saw signatures on page loads only, while scripts, beacons and other subresource requests from the same session went out unsigned, so a check that runs on every request will see a signed page followed by a string of unsigned calls from the same visitor. There is also no IP list to confirm the network, which leaves the signature as the only proof of identity.
The local modes are harder. When a user runs Manus through Browser Operator, the requests come from their real browser, cookies and home IP, and the only remaining clue is how the agent behaves on the page. The UC Davis researchers saw Manus jump the scroll position instantly, teleport the pointer to click targets and type far faster than a person.
How Verisoul detects Manus
Verisoul names Manus by verifying its Web Bot Auth signature against the key Manus publishes, which proves the request came from Manus and not from a scraper copying its user agent. We then tie that signed page load to the rest of the session, so unsigned scripts and beacons inherit the identity, and we check that the device and behavior match a cloud agent.
"agent": {
"id": "manus_bot",
"name": "Manus Bot",
"provider": "Manus",
"type": "ai_agent",
"detection_method": "signed",
"detection_signals": [
"cryptographic",
"behavioral"
]
}idis the stable name for the agent,manus_bot, which is what your rules should key ontypeisai_agent, meaning software acting for a person, as opposed to a crawlerdetection_methodissigned, meaning the identity was proven with a cryptographic signaturedetection_signalslists what backed it up: the verified signature, a cloud-hosted browser with a virtual machine device profile, and automated clicks, typing and page flow
The Agent Atlas entry for Manus Bot lists the same signals, and the AI agent detection docs cover the full agent object. Sessions from Browser Operator or the desktop app have no signature, so Verisoul flags them through bot score when the behavior is automated.
- 01User gives Manus a task
- 02Manus opens your site in its cloud browser
- 03Page loads are signed as api.manus.im
- 04Verisoul verifies it and returns manus_bot
How to detect Manus step by step
- Add the Verisoul SDK to the pages you care about, following the AI agent detection docs
- Read the
agentfield in each Verisoul authenticate response and logmanus_botsessions next to your own outcomes - Write rules on
agent.idand apply them to the whole session, since only page loads are signed - Send high bot scores with no agent name on sign-up and payout steps to review, which catches Browser Operator sessions
Should you block Manus?
For content and product pages, most teams let verified Manus sessions through or rate limit them, because the agent is usually doing research for a real user. Scraping is the main cost there, and rate limits on manus_bot handle it better than a block that also turns away paying users.
Sign-up deserves more care. Cue agents come with their own email addresses and phone numbers, so a sign-up from manus_bot can pass email and SMS checks without a person ever being involved. Checkout is usually fine when the account is established, while payouts, bonus claims and new payment methods are the steps to send to review.