Manus

TL;DR. Manus is an autonomous general-purpose AI agent from Butterfly Effect (Monica.im) that operates its own cloud virtual computer to plan and execute multi-step tasks, delivering finished reports, spreadsheets, websites, and code. Launched March 2025, it targets knowledge workers, startups, and small teams needing end-to-end task automation rather than conversational answers.

Overview

Manus is an autonomous general-purpose AI agent that operates its own virtual computer to independently plan, execute, and deliver complete work products such as reports, spreadsheets, websites, and code. Developed by Butterfly Effect (also known as Monica.im), the product launched in invitation-only beta on March 6, 2025, and was positioned as the world's first "general" AI agent, distinguishing itself from chatbots by taking action rather than merely answering questions.

Founding and leadership. Butterfly Effect was founded in 2022 by Xiao Hong (known as "Red"), a software engineer and Chinese entrepreneur who previously founded Nightingale Technology and created Monica, a ChatGPT-powered browser extension aggregating multiple LLMs. Co-founder Ji Yichao ("Peak Ji"), born in 1992 and named to MIT Technology Review's Innovators Under 35, serves as chief scientist and leads technical development. Zhang Tao, formerly head of global product at ByteDance, leads product. The company name "Manus" is derived from the Latin motto of MIT, "Mens et Manus" (mind and hand).

Funding and acquisition. Butterfly Effect raised a 2023 seed round led by ZhenFund, followed by a Series A in November 2024 bringing in HSG (formerly Sequoia China) and Tencent. In April 2025, the company closed a $75 million Series B led by Benchmark at a valuation of approximately $500 million, with participation from Tencent, HSG, ZhenFund, Wang Huiwen (Meituan co-founder), and Old Friendship Capital. On December 29, 2025, Meta announced its acquisition of Manus for over $2 billion, with plans to keep Manus operating as a standalone subscription product while integrating its technology into Meta's AI offerings. Xiao Hong will join Meta as a vice president.

Headquarters and team. Originally maintaining offices in Beijing and Wuhan, the company relocated its headquarters to Singapore in mid-2025, moving approximately 40 core technical staff and reducing its China-based workforce. As of late 2025, Manus employed about 105 people across Singapore, Tokyo, and San Francisco, with plans for a Paris office.

Revenue and scale. Manus reported surpassing $100 million in annual recurring revenue within eight months of launch, with a total revenue run rate exceeding $125 million. The company claimed to have processed over 147 trillion tokens and supported more than 80 million virtual computers. Microsoft began testing Manus in Windows 11 PCs in October 2025.

Key links: Website, Documentation, Blog.

Products & Features

Manus is a general-purpose autonomous AI agent that plans, executes, and delivers finished work across multi-step tasks. Developed by Butterfly Effect (branded as Monica) and launched in invitation-only beta on March 6, 2025, Manus gives an AI agent access to a full virtual computer (browser, terminal, file system) so it can complete real-world tasks autonomously, then continues processing in the cloud even when the user disconnects. Manus is now part of Meta.

Core capabilities. Manus accepts natural-language prompts and autonomously breaks them into sub-tasks, gathers information via a headless browser, writes code, creates files (PowerPoint, PDF, spreadsheets, websites, images), and delivers ready-to-use results. The agent runs in a cloud sandbox environment (Ubuntu, Python 3.10, Node.js 20) and supports task types including report writing, data analysis, app development, web design, slide creation, and content generation. Manus claims state-of-the-art performance on the GAIA benchmark across all three difficulty levels (86.5% on Level 1, 70.1% on Level 2, 57.7% on Level 3), reportedly surpassing OpenAI Deep Research.

Wide Research. A flagship feature that uses parallel multi-agent orchestration. A main agent breaks a request into hundreds of independent sub-tasks, each handled by a dedicated sub-agent with its own VM, tools, and fresh context. This avoids context-window saturation and scales to hundreds of items with uniform quality, delivering complete reports and datasets in minutes.

Browser Operator. A browser extension that works within the user's active browser session, using existing logins and tabs. It autonomously navigates, clicks, fills forms, and extracts data across sites without continuous supervision.

Toolkit and integrations. Additional tools include AI Design, AI Slides, AI Image Generator, and AI Music Generator. Mail Manus lets users trigger tasks via email. A Slack integration (with @manus mentions) enables task execution directly from team conversations. Team plans support SSO and admin controls.

API. A RESTful API (at open.manus.im) lets developers embed Manus's full agent capabilities (research, analysis, content creation, data processing) into their own applications, returning complete deliverables rather than just text responses.

Platform availability. Web app, desktop app (macOS and Windows, with a "My Computer" feature for local file management and command execution), and mobile app (iOS 17.0+ and Android 7.0+).

Pricing. Manus uses a credit-based model where credits are consumed based on task complexity. Plans include Free ($0, with 300 daily refreshed credits), Standard ($20/month, 4,000 monthly credits), Customizable ($40/month, 8,000 credits), and Extended ($200/month, 40,000 credits). Team plans with shared credit pools and enterprise sales are also available. Credits can be supplemented with add-on packs that never expire.

Security & Compliance

Manus advertises an enterprise-grade security posture with SOC 2 Type II, SOC 2 Type I, ISO 27001:2022, and ISO 27701:2019 certifications, alongside significant geopolitical scrutiny tied to its Chinese origins. The official Manus security page lists these four certifications, with attestation letters and the ISO 27001 certificate available via the Manus Trust Center. Enterprise features include SSO integration, granular sharing permissions, audit logs, role-based access controls, and a zero-trust sandbox architecture where each task runs in an isolated cloud virtual machine that cannot access other sessions or account data.

Manus states it does not train models on Team or Enterprise customer data and explicitly prohibits model providers from training on that data. Customer data is deleted upon account termination. Data encryption is utilized in transit and at rest, with restricted encryption key access. The sandbox follows a zero-trust principle, and connectors are automatically disabled when collaboration mode is enabled to prevent collaborator access to integrated services. Manus is not HIPAA compliant and will not sign a Business Associate Agreement; its Terms of Service prohibit users from submitting protected health information.

Despite these certifications, Manus faces significant security and data sovereignty concerns. The platform is operated by Butterfly Effect Pte. Ltd., a Singapore-domiciled entity, but reporting from the South China Morning Post, Forbes, and others confirms the team is based in Beijing and Wuhan, raising questions about potential data transfers to China under Chinese law. In March 2025, Tennessee became the first U.S. state to ban Manus from government networks, citing data privacy and cybersecurity threats from platforms with ties to the Chinese Communist Party. The privacy policy, governed by Singaporean law, has been criticized by legal analysts for appearing AI-generated and containing generic GDPR explanatory sections rather than precise data handling practices. No publicly confirmed data breaches have been reported, though the expanded attack surface of autonomous agents that log into third-party services and store session credentials remains an inherent risk.

User Feedback & Adoption

Manus AI generated massive initial hype but user sentiment is sharply polarized between enthusiastic early adopters and frustrated paying customers. The product went viral at launch in March 2025, with over 2 million people joining the waitlist within the first week and the official Discord community growing past 138,000 members within days. Invite codes reportedly resold on Chinese secondary markets for as much as approximately USD 7,000. Andreessen Horowitz named Manus among its consumer AI "breakout" products of 2025, and Sacra estimates the company reached roughly $127M in annualized revenue by December 2025, climbing toward an estimated $450M by June 2026.

Review ratings reflect the polarization. On G2, Manus AI Agent holds 2.2 stars from 5 verified reviews, one of the lowest ratings on the platform. The Apple App Store tells a different story: 4.7 stars from approximately 35,000 ratings, though these skew toward casual and mobile-first users. AppBrain reports 7.6 million Google Play downloads with a 4.54-star average. Trustpilot reviews for manus.im are predominantly negative, with users citing unauthorized credit consumption, poor customer service, and difficulty canceling subscriptions.

Common praise themes center on Manus's autonomous web research, data extraction into structured spreadsheets, and website and app building for prototypes and MVPs. Users appreciate the asynchronous cloud execution model and the breadth of tool integrations. Common complaint themes are consistent across G2, Trustpilot, Reddit, and independent reviewers: unpredictable credit burn (with users reporting a single complex task consuming 900+ credits), instability and crashes under load, tasks getting stuck on paywalls or CAPTCHAs, generated code requiring significant debugging, and credit-consuming loops where the agent repeats mistakes. Multiple reviewers note a sharp degradation in reliability as task complexity increases, with the agent excelling at standard patterns but struggling with novel business logic.

Notable adoption milestones include a partnership as a launch partner for Microsoft Agent 365 (November 2025) and Meta's reported acquisition at a valuation exceeding $2 billion (December 2025), though China blocked the acquisition in April 2026. Manus 1.5 (October 2025) reduced average task completion time from 15 minutes to under 4 minutes and the company reported a 6% user satisfaction increase, while Manus 1.6 Max (December 2025) reportedly raised satisfaction by over 19% in double-blind testing.

Monetization & Business Model

Manus operates a freemium SaaS model with credit-based usage pricing, targeting individual professionals, teams, and enterprises. The company introduced its first paid subscription plans in March 2025, transitioning from a free, invite-only beta that had accumulated a waitlist exceeding 2 million users.

Pricing Tiers (current, per official Manus Help Center):

• Free: $0/month with 300 daily refresh credits, access to Manus 1.6 Lite in Agent Mode and Chat Mode.

• Pro (Tier 1): Starting at $20/month with 4,000 monthly credits (17% discount with annual billing).

• Pro (Tier 2): Starting at $40/month with 8,000 monthly credits, includes a 7-day free trial (17% discount with annual billing).

• Team: Starting at $20 per seat/month (17% discount with annual billing).

Earlier pricing (March 2025 launch) offered a Starter tier at $39/month with 3,900 credits and a Pro tier at $199/month with 19,900 credits, per Sacra. The pricing structure has since been revised downward. Credits are consumed per task based on complexity, and annual billing offers a 17% discount across all paid plans. An enterprise SDK is under development to allow third-party embedding of Manus workflows.

Funding & Valuation: Manus (parent company Butterfly Effect) raised a $75 million Series B in April 2025 led by Benchmark at a ~$500 million valuation, roughly quintupling its prior valuation. Total disclosed funding is approximately $85 million, including an earlier $10 million+ Series A. Investors include Benchmark, Tencent, HongShan (formerly Sequoia China), ZhenFund, Wang Huiwen (Meituan co-founder), and Old Friendship Capital. Benchmark general partner Chetan Puttagunta joined the board.

Revenue & Trajectory: Sacra estimates Manus hit approximately $127 million in annualized revenue in 2025, climbing to ~$450 million annualized by June 2026, with over 20% month-over-month growth following the Manus 1.5 release. The company had roughly 105 employees across Singapore, Tokyo, and San Francisco as of December 2025.

Meta Acquisition (December 2025): Meta announced its acquisition of Manus in December 2025 at a reported valuation of $2-3 billion. China's government blocked the deal in April 2026 citing national security concerns. Meta completed an operational split by June 2026, and Manus's Chinese venture backers were reportedly pursuing a buyback at roughly the same $2 billion valuation, potentially positioning the company for a Hong Kong IPO.

Leadership & Team

Manus is led by three co-founders: CEO Xiao Hong (Red), Chief Scientist Ji Yichao (Peak), and Chief Product Officer Zhang Tao (Hidecloud). All three relocated from China to Singapore in mid-2025 as part of the company's headquarters move, and all joined Meta in December 2025 when the social media giant acquired Manus for a reported $2 billion+.

Board and investors: Benchmark general partner Chetan Puttagunta joined the board following the April 2025 Series B round. Notable investors include Benchmark (Series B lead, $75M at ~$500M valuation), Tencent, HSG (formerly Sequoia China), ZhenFund, and Wang Huiwen (Meituan co-founder).

Team size and trajectory: The company employed approximately 120 staff in China before relocating its headquarters to Singapore in mid-2025. Around 40 core technical staff relocated; the remainder were laid off. By late 2025, Manus reported about 105 employees across Singapore, Tokyo, and San Francisco, with plans for a Paris office. Following the Meta acquisition, Manus employees joined Meta's teams, with Xiao stating the deal would not change how Manus operates or how decisions are made.

Target Audience & Use Cases

Manus targets knowledge workers, startups, and business teams who need autonomous end-to-end task execution rather than conversational AI assistance. Its core audience is professionals and small teams whose work involves complex, multi-step knowledge tasks where high autonomy delivers significant time savings.

Primary market segments. Manus serves individual professionals on free and Pro plans ($0 to $200/month), teams on per-seat Team plans (starting at $20/seat/month with shared credit pools), and early-stage startups through a dedicated Manus for Startups program. Enterprise features include SSO (via WorkOS), data training opt-out, usage analytics, and access controls. The startup program offers 6 months of Team subscription for up to 20 seats (valued at $4,800) for companies under 5 years old that have raised $500K to $50M in equity.

Target personas. Financial analysts use Manus for stock market analysis, investor reports, and financial modeling. Market researchers and consultants leverage it for competitive analysis, supplier sourcing, and B2B lead generation. Developers and technical founders use it for prototyping, code generation, and full-stack application building. Marketing professionals deploy it for content briefs, campaign research, and customer prospect mapping. Educators and individual users apply it to curriculum development, travel planning, and product research.

Top use cases. (1) Research and data analysis: market research, competitive analysis, and structured data extraction from databases like Y Combinator company lists. (2) Financial analysis and reporting: stock performance analysis with interactive dashboards, correlation studies, and investor reports. (3) B2B lead generation and supplier sourcing: identifying prospects and evaluating suppliers across extensive networks. (4) Content creation and educational materials: generating interactive courses, video presentation scripts, and visualization maps. (5) Business operations: sales data analysis, contract review, and recruitment scheduling.

Industry verticals. E-commerce and retail (sales analytics, store operation analysis), finance (stock analysis, forecasting), education (curriculum and learning materials), and B2B SaaS (lead gen, competitive intelligence).

Company size sweet spot. Small teams and startups are the core fit. Manus explicitly positions that a team of 5 can operate like 15 with its agents, targeting lean organizations competing with larger teams. The Team plan with shared credits and SSO addresses growing organizations, though enterprise-grade governance remains limited.

Geographic focus. Manus supports 16 interface languages including English, Chinese, Japanese, Korean, French, German, Spanish, Portuguese, Arabic, Hindi, and Thai, indicating a global rather than region-specific focus. The product was originally developed by the Chinese startup Monica.im.

Tags & Categories

Primary Category: AI Other Categories: AI Other, AI Coding, AI Analytics, AI Writing, AI Design Tags: autonomous ai agent, general-purpose agent, task automation, web browsing, workflow automation, code generation, data analysis, research, sandbox environment, cloud-based, meta, slack integration

Manus is an autonomous general-purpose AI agent that executes multi-step tasks end to end rather than answering questions conversationally. Its official documentation describes it as "an autonomous general AI agent designed to complete tasks and deliver results," operating in a sandboxed virtual computer with internet access, a persistent file system, and the ability to install software and create custom tools. This breadth of capability, spanning coding, data analysis, research, design, and content creation, places it outside any single specialized category and into the general "AI Other" bucket, with secondary overlap in AI Coding, AI Analytics, AI Writing, and AI Design.

Founded by the team at Butterfly Effect (later acquired by Meta in 2026), Manus launched in invitation-only beta on March 6, 2025. It distinguished itself early by topping the GAIA benchmark, a general-purpose AI agent evaluation, scoring 86.5% on Level 1 tasks versus OpenAI Deep Research's 74.3%. The product's name derives from the Latin phrase "Mens et Manus" (mind and hand), reflecting its dual emphasis on reasoning and action.

Manus is available via web app, desktop, and mobile, with integrations including Slack and an API. Its feature set includes Wide Research (parallel multi-agent research), a browser operator, and tools for generating slides, websites, images, and music. Pricing tiers include Basic, Plus, and Pro, with team plans offering SSO.

Impact & Recommendations

Manus went from viral launch to geopolitical flashpoint in under 14 months, illustrating both the surging demand for autonomous AI agents and the acute regulatory risks facing Chinese-origin technology. Butterfly Effect launched Manus in March 2025 to explosive interest: its demo video surpassed 1 million views in under 20 hours, Discord topped 138,000 members within days, and invite codes resold for up to $13,800 on secondary markets. By December 2025, the company reported over $100 million in annualized revenue, eight months post-launch, with a run rate exceeding $125 million. Meta acquired Manus for over $2 billion, but in April 2026 China's NDRC ordered the deal unwound, blocking the transfer of Chinese-origin AI technology to a U.S. buyer and leaving Manus in operational limbo.

Competitive landscape. Manus positioned itself as a general-purpose autonomous agent, claiming state-of-the-art scores on the GAIA benchmark (Level 1: 86.5%, Level 2: 70.1%, Level 3: 57.7%), surpassing OpenAI Deep Research (67.36%). Its closest competitors are OpenAI Operator (now ChatGPT Agent), Anthropic Claude Computer Use, and open-source frameworks like Browser Use. Manus differentiates through its multi-agent architecture (executor, planner, knowledge agents), sandboxed Linux environment with code execution, and direct web deployment via Manus Spaces. However, critics noted Manus relies on Anthropic's Claude and Alibaba's Qwen rather than proprietary foundation models, raising questions about its defensibility as a "wrapper."

Growth signals. $75 million Series B led by Benchmark (April 2025) at a ~$500 million valuation, backing from Tencent and HSG, over $100M ARR within eight months, Microsoft testing Manus in Windows 11 (October 2025), and Meta's willingness to pay over $2B all signal strong market traction.

Risk factors. Execution reliability remains a core concern: MIT Technology Review found Manus "promising but not perfect," with frequent crashes, CAPTCHA blocks, and tasks taking hours. Reviewers reported endless loops, context overflow, and output quality inconsistent for production use. The NDRC's blocking of the Meta acquisition creates profound uncertainty around Manus's ownership structure, Chinese asset obligations, and ability to retain relocated talent in Singapore. Data privacy concerns persist given the platform's closed-source nature and Chinese origins.

ICP fit. Manus best serves knowledge workers and teams needing multi-step research, data analysis, and deliverable generation (reports, dashboards, websites) on the open web. Pricing starts at $0 (300 daily credits) with paid tiers from $20/month (4,000 credits) to $200/month (40,000 credits). The credit-based model makes costs unpredictable for heavy usage. For the Chiri Atlas audience of AI-savvy builders and operators, Manus is a compelling tool for exploratory, non-mission-critical autonomous workflows, but the ownership upheaval makes it a high-risk bet for production dependencies.

Outlook. Manus demonstrated that autonomous agents can achieve real revenue and user demand at scale, but its trajectory also shows that geopolitics, not just technology, now determines which AI tools survive. The unwound Meta deal leaves Manus without a clear strategic backer at a critical growth stage, and its reliance on third-party foundation models limits its moat. Competitors with integrated model stacks (OpenAI, Anthropic, Google) are better positioned for durable advantage, though Manus's head start in multi-agent orchestration remains a genuine differentiator if it can stabilize operationally and resolve its ownership uncertainty.

Chiri Analysis

Chiri Score: 72/100

Verdict

Best for:

• Knowledge workers needing autonomous multi-step research and structured data extraction

• Startups and lean teams wanting agent leverage (the startup program offers 6 months for up to 20 seats)

• Financial analysts and consultants building reports, dashboards, and competitive analyses

• Developers and technical founders prototyping apps, code, and MVPs

• Teams running exploratory, non-mission-critical autonomous workflows

Not for:

• Organizations requiring HIPAA compliance or a signed BAA (explicitly unsupported)

• U.S. government agencies and entities with China data-sovereignty restrictions (banned in Tennessee)

• Production-critical workflows needing predictable costs and high execution reliability

• Buyers needing ownership and vendor stability certainty after the blocked Meta acquisition

• Heavy-usage users sensitive to unpredictable credit consumption

Head-to-head

Frequently Asked Questions

Is Manus SOC 2 compliant?

Yes. Manus holds SOC 2 Type II, SOC 2 Type I, ISO 27001:2022, and ISO 27701:2019 certifications, with attestation letters available through the Manus Trust Center.

How much does Manus cost?

Manus uses credit-based pricing with a Free plan at $0/month (300 daily refresh credits), Pro Tier 1 at $20/month (4,000 credits), Pro Tier 2 at $40/month (8,000 credits with a 7-day trial), and an Extended plan at $200/month (40,000 credits). Team plans start at $20 per seat/month, with a 17% annual billing discount.

Who are Manus's competitors?

Manus competes with OpenAI Operator (ChatGPT Agent), Anthropic Claude Computer Use, OpenAI Deep Research, and open-source frameworks like Browser Use. It differentiates through multi-agent orchestration and a persistent sandboxed virtual computer.

Is Manus good for enterprise?

Manus offers enterprise features including SSO, audit logs, role-based access controls, and a zero-trust sandbox, but execution reliability issues, unpredictable credit costs, no HIPAA support, and post-acquisition ownership uncertainty limit its fit for mission-critical production use.

Did Meta acquire Manus?

Meta announced acquiring Manus for a reported $2 billion+ in December 2025, but China's government blocked the deal in April 2026 over national security concerns, leaving Manus's ownership structure unresolved.

What is Manus and how does it work?

Manus is an autonomous general-purpose AI agent that operates its own cloud virtual computer, browser, terminal, and file system, to independently plan, execute, and deliver complete work products. It breaks natural-language prompts into sub-tasks and continues processing in the cloud even after the user disconnects.

Does Manus train on customer data?

Manus states it does not train models on Team or Enterprise customer data and prohibits model providers from training on that data. Customer data is deleted upon account termination, with encryption in transit and at rest.

How does Manus perform on the GAIA benchmark?

Manus claims state-of-the-art GAIA performance of 86.5% on Level 1, 70.1% on Level 2, and 57.7% on Level 3, surpassing OpenAI Deep Research's 74.3% on Level 1.

---

Reviewed by Chiri Atlas Research Desk (AI Tooling Analyst) on 2026-07-08.