Manus
Also known as: Manus AI, Butterfly Effect
Autonomous general AI agent that takes a goal, plans and executes the steps in its own cloud computer and browser or in your local browser, and returns finished research, slides, spreadsheets and web apps, with scheduled tasks, email triggers, connectors and an API.
Manus is an autonomous, general purpose AI agent. You give it an objective in plain language and it plans the steps, carries them out, and returns a finished result: research reports, spreadsheets, slides, images and working web apps with a backend, database and hosting. It works in its own cloud computer and cloud browser, and the Browser Operator extension lets it navigate, click and fill forms inside the user's own browser with existing logins once the user grants control. Wide Research runs many sub agents in parallel for large research jobs.
Work reaches Manus from the app, from Scheduled Tasks that run on a recurring schedule, from Mail Manus bot addresses that turn forwarded email into tasks, from Slack and from a REST API with webhooks.
Projects keep a master instruction and a knowledge base of files that every task in the project inherits, a library of ready to use Skills built by Manus and the community packages repeatable workflows, and MCP connectors plus custom MCP servers connect it to Gmail, Calendar, Notion, GitHub and internal tools.
The agent pauses for confirmation before supported actions such as sending an email, and Plan Mode lets the user approve the approach before execution. The Team plan adds SSO, admin access controls, audit logs and pooled credits, and Manus holds SOC 2 Type II and ISO 27001.
Meta acquired Manus on 29 December 2025. The acquisition was unwound on regulatory grounds, affected users' data from after that date was deleted and restored in a separation on 23 to 24 August 2026, and Manus has since formally resumed independent operations, led by its founding team, with data stored in the US and Singapore. It fits researchers, solo builders and teams that want a hands off agent for open ended tasks, and is a weaker fit for teams that need self hosting, a choice of model provider, or predictable per task cost.
Vendor details
Canonical URL
https://manus.im
Category
Browser / computer-use agent
Subcategory
Autonomous general agent in a cloud virtual computer
Funding status
Independent. Meta acquired Manus on 29 December 2025; the deal was unwound on regulatory grounds and Manus formally resumed independent operations after a data separation on 23 to 24 August 2026. Earlier funding included a 2023 seed led by ZhenFund, a Series A in November 2024 with HSG and Tencent, and a Benchmark led Series B in April 2025.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Operates Manus's own cloud computer and browser, or the user's local browser through the Browser Operator extension; MCP connectors (Gmail, Google Calendar, Notion, GitHub, Slack and others), custom MCP servers for internal systems, Zapier, and built in data sources such as Similarweb; a documented REST API (API key or OAuth2) with webhooks, projects, files and agents.
In practice
A researcher gives Manus a one sentence brief to compare the top ten agent platforms on pricing and build a spreadsheet. It browses each site, extracts the data, runs code, and returns the finished sheet without further input.
A builder asks for a working prototype. Manus builds a working web app with a backend, a database and hosting from a single prompt.
An analyst needs a deep dossier. Wide Research runs many sub agents in parallel across sources and assembles a research report while the user is away.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
10.0 / 14 capabilities · 71%
| Integrations & Tool Calling | Full |
|---|---|
|
A catalog of MCP connectors (Gmail, Google Calendar, Notion, Slack, GitHub and others) lets the agent act in outside systems. Custom MCP servers add internal tools and proprietary APIs, and Zapier reaches further apps. In the API, connectors can be chosen per task. Sourcemanus.im/docs/integrations/custom-mcpread 2026-09-27 |
|
| Workflow Orchestration | Full |
|
The agent plans and executes multi step tasks end to end, Wide Research runs a parallel multi agent system across many sub tasks, and custom agents run subtasks listed through the API. Sourcemanus.im/docs/features/wide-researchread 2026-09-27 |
|
| Knowledge Grounding & RAG | Partial |
|
Projects keep a knowledge base of uploaded files that every new task in the project inherits, and connectors pull from Notion, Drive and similar sources per run. Manus does not describe an index, embeddings layer or retrieval structure over that knowledge. Web research gathers context for each task as it runs. Sourcemanus.im/docs/features/projectsread 2026-09-27 |
|
| Human Oversight & Guardrails | Full |
|
The agent pauses and waits for the user to confirm supported actions, such as sending an email, before it goes on. The user answers in the app or through task.confirmAction. Plan Mode shows the approach as an editable document, and the user approves it before work starts. Sourceopen.manus.im/docs/v2/task-lifecycleread 2026-09-27 |
|
| Security, Identity & Governance | Full |
|
The Team plan sets up SSO with the customer's own identity provider after domain verification. It has Team Owner and Super Admin roles and an admin dashboard with access control and granular sharing permissions. Manus holds SOC 2 Type II and ISO 27001, the SOC 2 attestation letters and the ISO certificate are listed on the security page, and a trust center sits at trust.manus.im. Sourcemanus.im/teamread 2026-09-27 |
|
| Observability & Auditability | Full |
|
Each task keeps a full event history of the agent's steps, readable in the app and through task.listMessages. The Team plan adds audit logs and an admin dashboard with usage tracking. Usage history is also shown per session and per member. Sourcemanus.im/teamread 2026-09-27 |
|
| Memory & State Persistence | Partial |
|
Projects keep a master instruction and files that users maintain and every new task reads, so users keep this store and it carries across sessions. Stopped tasks resume with their state. There is no memory layer that the agent writes itself, with a set scope and lifetime. Sourcemanus.im/docs/features/projectsread 2026-09-27 |
|
| Deployment & Data Residency | Not documented |
|
Manus runs only in its own cloud. Its data is stored in the US and Singapore, and the customer cannot choose a region. There is no self hosted, VPC or residency option. Sourcemanus.im/blog/a-note-to-our-usersread 2026-09-27 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
A categorized library of ready to use Skills, built by Manus and the community, includes Excel Generator, Financial Analysis, SEO Competitor Analysis and Listicle Blog Writer. Skills are added to a workspace from the official library, uploaded, or imported from GitHub, and each Skill does its own job. Sourcemanus.im/features/skillsread 2026-09-27 |
|
| Triggers & Channel Coverage | Full |
|
Scheduled Tasks run work on a recurring schedule or at a set time (20 per paid plan), Mail Manus starts tasks from forwarded or CC'd email including dedicated workflow addresses, and tasks also arrive from Slack and the API with webhooks for task events. Sourcemanus.im/docs/features/scheduled-tasksread 2026-09-27 |
|
| Model Flexibility & Routing | Not documented |
|
The customer picks one of three agent profiles, standard, lite or max, and model versions cannot be chosen separately through the API. These are tiers of Manus's own agent, and Manus does not name the underlying model provider. Sourceopen.manus.im/docs/v2/task.createread 2026-09-27 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The Manus API is a REST API with API key or OAuth2 authentication and published OpenAPI specifications. It creates, continues, stops and lists tasks, manages projects, files, webhooks and agents, and publishes websites built by the agent. Sourceopen.manus.im/docs/v2/introductionread 2026-09-27 |
|
| Testing, Debugging & Optimization | Not documented |
|
The Skill Creator skill is a guided workflow for authoring Skills, and it includes a testing step. It does not report a verdict on agent behavior. There is no evaluation harness, scored test case or quality gate for the agent. Sourcemanus.im/docs/llms.txtread 2026-09-27 |
|
| Browser & Computer Use | Full |
|
The agent drives Manus's own cloud browser, which operates like a real person. The Browser Operator extension lets it navigate, click and fill forms in the user's local browser with existing logins, once the user grants control. Manus also has a desktop app. Sourcemanus.im/docs/features/browser-operatorread 2026-09-27 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Manus 2.0 shipped on web, desktop and mobile, built on Cascade, a rebuilt agent harness that loads specialized capabilities only when a project needs them. A new desktop workspace, Manus Studio, lets the agent work directly with approved local files and tools, and Remote Control lets a voice command from a phone have Manus operate the desktop while the screen streams live. Scheduled tasks also became Automations that can start when something happens in a connected app, such as a new pull request or a refund.
Bears on: Browser/computer use
View sourceManus announced its return to independent operations after regulators ordered the unwinding of its December 2025 acquisition by Meta. As part of this separation, data generated by affected users on or after the acquisition date will be permanently deleted between August 23 and August 24, 2026. Affected accounts will temporarily lose service during this transition. Users must export their workflows using the provided backup tool before August 23 and restore them when the portal opens on August 25.
Bears on: Deployment / data residency
View sourceManus launched Plan Mode, an opt-in feature that adds a human-in-the-loop review step before the AI agent executes a build or task. When activated, the agent generates an editable Markdown document detailing its planned goals, constraints, and implementation steps. Execution is paused and nothing is built until the user confirms or modifies the proposed approach.
Bears on: Human approval / guardrails
View sourcePricing
Free account (lite agent); paid from $20 a month for 4,000 credits, $40 for 8,000 and $200 for 40,000, each with 300 daily refresh credits; Team $20 a seat a month
credit based subscription; every action consumes credits, complex tasks consume many
What is public
All individual tiers and the Team seat price are public, monthly and annual, with credit allowances stated; credit consumption per task is not fixed.
Billing mechanics
Credit based subscription tiers with a monthly allowance plus 300 daily refresh credits; add on credits never expire (Team add ons stay valid while the subscription is active); Team pools each seat's 4,000 credits across members, and seats adjust at the next billing cycle; annual billing saves 17 percent.
Cost watchouts
Every task consumes credits in proportion to its complexity, so long running or Wide Research tasks can use a month's allowance quickly; buying add on credits or moving up a tier is how heavy use is covered.
Variable cost rationale
Pricing is credit based and every action consumes credits, with complex tasks reported to burn four thousand to ten thousand in a single run and monthly credits not rolling over, so cost can escalate fast and unpredictably with usage.
Additional watchouts
Because pricing is credit based and consumption scales with task complexity, heavy or long running agent workflows can escalate cost quickly; budget against expected task volume, not just the sticker tier price.
Sales call required
No, self serve available
Free / trial
Free personal accounts exist and run the lite agent profile; the credit allowance for free accounts is not stated on the pricing page
Lowest paid plan
$20 a month for 4,000 monthly credits plus 300 daily refresh credits; $40 and $200 tiers for 8,000 and 40,000
Key ambiguities
How many credits a given real workload will consume, since consumption scales with task complexity and long running agents are the biggest cost driver.
Missing data
Free account credit allowance; any enterprise tier above Team is sold through contact sales without a published price.
Related vendors
- AgentQL — A self-healing query language for the web: natural language…
- AGI, Inc. — AGI, Inc
- Airtop — Cloud browsers for AI agents that compile plain English workflows…
- Asteroid — Healthcare portal integration platform: supervised browser and…
- Automat AI — AI agents that operate computers visually, sold as Automat Workforce…
- Autotab — General AI agent that learns a workflow from a demonstration and…
Alternatives to Manus
The closest documented capability profiles to Manus among browser and computer-use agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Asteroid11.5 / 14Adds documented Model Flexibility & Routing and Testing, Debugging & Optimization
- Fellou8.0 / 14A lighter documented profile than Manus
- Automat AI11.5 / 14Adds documented Deployment & Data Residency and Model Flexibility & Routing, among others
- Perplexity Comet9.5 / 14Adds documented Model Flexibility & RoutingManus vs Perplexity Comet →
- Airtop9.0 / 14Adds documented Model Flexibility & Routing and Testing, Debugging & Optimization
- Claude in Chrome9.0 / 14Adds documented Deployment & Data Residency
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded