Manus

What it is

Manus is a general-purpose autonomous agent. You give it a goal — “build a list of every mid-market logistics company that changed CFO this year, with sources” — and it plans the steps, opens a browser on its own cloud computer, writes and runs code, and hands back a finished file: a spreadsheet, a slide deck, a report, or a deployed web page. Tasks keep running when your laptop is closed, and up to 20 can run at once on any paid plan.

The company behind it, Butterfly Effect, launched Manus in March 2025 and moved its headquarters from China to Singapore that summer. Meta bought it for roughly $2 billion in December 2025. China’s National Development and Reform Commission ordered the deal withdrawn on 27 April 2026, Meta cut ties in June, and Manus resumed independent operations on 1 September 2026 after investors led by Tencent, ZhenFund, and HSG bought the shares back from Meta. Tencent is now its largest shareholder. Caixin reports annual recurring revenue of $400 million by late June 2026, a figure the company has not had independently audited.

Manus 2.0, released 29 September 2026, adds trigger-based tasks (an email, a Slack message, or a calendar event can start a run), an always-on cloud computer, and remote control of your own desktop from your phone. A companion product, Cue, gives agents their own email address, phone number, and spending wallet; it is invite-only.

Why it lands on shortlists

• It finishes the deliverable. Most assistants stop at an answer. Manus produces the artifact — a cleaned account list, a 15-slide competitor brief, a data-room index — which is what an ops team actually needs to hand to someone else.
• Wide Research runs in parallel. For list-building work (100 target accounts, 40 vendor security pages, 60 candidate profiles), Manus fans out sub-agents that each take one item, then merges the results. That is the job a single chat thread does badly.
• Scheduled and triggered runs. Each paid plan allows 20 scheduled tasks, and 2.0 adds event triggers from Gmail, Slack, Google Calendar, and Notion. A weekly competitor-pricing sweep or a Monday pipeline-risk digest runs without anyone opening the app.
• Extensible enough for ops. Prebuilt connectors cover GitHub, Notion, Slack, and Google Calendar; custom MCP servers connect it to internal systems; and an API lets you start tasks from your own code or an automation tool.

What it costs

Individual plans on the official pricing page, checked 2 October 2026: $20 a month for 4,000 credits, $40 for 8,000, and $200 for 40,000, each plus 300 refresh credits a day that do not carry over. Annual billing saves 17%. The Team plan is $20 per seat per month (or $200 per seat per year) with 4,000 credits per seat in a shared pool, plus SSO, an admin dashboard, and audit logs; Manus states SOC 2 Type II and ISO 27001 compliance.

Credits are the real price, and they are hard to predict. Manus’s own help-center examples put a 15-minute data-analysis chart at 200 credits, a 25-minute web page at 360, and an 80-minute web app at 900. Credits are consumed by model tokens, the virtual machine, and third-party API calls, and the product does not quote a cost before a run. By those examples, a 4,000-credit seat covers somewhere between 4 and 20 tasks a month, depending on how large they are. A five-person ops team doing weekly list-building lands on the $40 tier or buys add-on credits, so budget $40-$60 per active user per month, not $20.

Best for

• RevOps and marketing-ops leads who build target-account lists, competitor briefs, and territory research by hand and want the finished spreadsheet, not a chat transcript
• Legal-ops and procurement analysts running repeatable desk research — vendor security pages, public filings, regulatory trackers — on a schedule
• Recruiting ops teams sourcing market maps and compensation benchmarks from public data

It is a poor fit for anything that must touch privileged documents or regulated personal data until your security team has signed off on the ownership change (see watch-outs).

Alternatives — and when to pick them instead

• ChatGPT agent mode and Claude Cowork — the market-share leaders, and usually already licensed. Pick them when the task is under 30 minutes, the work lives in files you already have, or procurement will not add a new vendor. Cowork’s advantage is running against your local files; Manus’s is a long, parallel, browser-heavy run in the cloud.
• Perplexity — pick it when you need a cited answer, not a built artifact. It is faster and cheaper per question.
• Lindy — pick it when the job is a standing workflow (triage an inbox, update the CRM on every call) rather than a one-off research project. Its agents are built around triggers and integrations first.
• OpenClaw — pick it when data must stay on infrastructure you control. Self-hosting costs engineering time but removes the vendor-continuity question entirely.
• Manus is itself the fastest-growing entrant in this group: from launch to a reported $400 million in annual recurring revenue in about 15 months.

Watch-outs

• Ownership has changed twice in nine months. The Meta acquisition was unwound by a Chinese regulator after the company had relocated to Singapore, and Tencent is now the largest shareholder. Data is stored in the US and Singapore, but the episode shows that Chinese regulators can still reach the company. Guard: run a fresh vendor-risk review against the post-September entity before standardizing, keep a second agent (Cowork or ChatGPT) licensed, and do not route customer PII or privileged legal material through it until that review closes.
• Manus has already deleted user data once. On 23-24 August 2026 it deleted data created after 29 December 2025 for affected accounts, citing unnamed jurisdictional requirements; the export window closed at 07:59 Singapore time on 23 August. Guard: treat Manus as a workspace, not a system of record — export every deliverable to your own drive or CRM the day it is produced, and schedule a monthly export of task history.
• Credit spend is invisible until after the run. A broad Wide Research task can consume a large share of a monthly allowance, and nothing warns you first. Guard: pilot on the $20 tier for two weeks, log credits per task type from the usage dashboard, and set auto-purchase limits on the Team plan before rolling it out.
• Autonomy means unreviewed actions. An agent with a browser, connectors, and (with Cue) its own wallet can send, submit, or spend. Guard: connect read-only scopes first, keep write access to the CRM and email off until outputs have been reviewed for a few weeks, and do not enable Cue spending for business use.