Delv
General Assistantby Butterfly Effect★ 4.3

Manus

General-purpose autonomous agent that can browse the web, write code, and run multi-step workflows. The viral "show me everything" agent.

Try it Also in Tools: AI Code & Dev →
C
Safety & Trust

Delv Safety Grade: C

Score 58/100 · assessed 2026-04-18

Maintainer45
Permissions35
Supply chain40
Transparency30
Incidents100

Manus is an autonomous agent from Butterfly Effect that gained viral attention for multi-step workflow execution including web browsing and code generation. The maintainer appears to be a smaller organisation with limited public track record. Major concerns centre on the closed-source nature (no public repository), lack of transparency about implementation details, and extremely broad permissions required for autonomous web browsing, code execution, and workflow orchestration. The freemium model suggests a hosted service, which raises questions about data handling and execution environment. Supply chain is opaque with no clear distribution mechanism beyond their website. No known security incidents, but the combination of autonomy, broad capabilities, and limited transparency creates meaningful risk for sensitive workflows. Suitable for non-sensitive research tasks but requires careful consideration for anything involving proprietary data or credentials.

Green flags

  • No known security incidents or credential leaks to date
  • Viral adoption suggests real-world utility and user validation
  • Freemium model allows testing before committing to paid tier
  • Documented use case for competitive research demonstrates practical value

Red flags

  • No public repository or source code available for audit
  • Autonomous web browsing with unclear sandboxing or boundary enforcement
  • Code execution capabilities without documented safety constraints
  • Unknown data retention and privacy practices for hosted service
  • Limited information about Butterfly Effect as maintainer organisation

Permissions requested

Outbound networkBrowser controlShell executeWrite filesExternal LLM callRead env
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Pricing

FREEMIUM

Platforms

webapi

Review

Manus went viral for a reason: it's the first autonomous agent that feels like it actually gets work done without constant babysitting. I've used it for competitive research where I needed pricing, feature matrices, and customer sentiment from a dozen SaaS tools. Instead of opening twenty tabs and copy-pasting into a spreadsheet, I gave Manus the company names and asked for a comparison table. It browsed each site, extracted structured data, cross-referenced reviews, and returned a usable CSV in about eight minutes. The autonomy here is real. Manus plans multi-step workflows, backtracks when it hits dead ends, and doesn't need you to prompt it through every click. For research tasks that involve synthesising information from scattered sources, it's genuinely faster than doing it yourself. The "show me everything" demo that made the rounds on Twitter wasn't a party trick - that's how it works. You describe an outcome, it figures out the steps. Where it shines: open-ended research, purchase decisions with trade-offs ("find me the best project management tool for a remote team under 20 people"), and report drafting that needs live data. I've also used it to draft technical documentation by pulling API specs and example code from GitHub repos. It's not perfect - sometimes it misinterprets a site's structure or fetches stale data - but it's right often enough that I reach for it first. Failure modes: it struggles with tasks that need domain expertise to evaluate quality. If you ask it to "find the best React component library", it'll give you a list based on GitHub stars and npm downloads, but it won't know that one has terrible TypeScript support or that another ships 400KB of unshakeable CSS. It's also slow when the web is involved - expect minutes, not seconds. The free tier is generous but rate-limited; you'll hit the cap quickly if you're using it daily. Compared to AutoGPT or similar open-source agents, Manus is far more reliable. It doesn't spiral into recursive loops or hallucinate file paths. Compared to Devin (for code) or Perplexity Pro (for research), it's broader but shallower. If you need one tool that does both browsing and coding adequately, Manus is the pick. If you need deep expertise in either, specialise.
Verdict

Pay for Pro if you do regular competitive research, market analysis, or purchase decisions that involve comparing five-plus options. Skip it if you need domain-specific expertise or can't tolerate the occasional wrong turn.

Good at

  • Actually autonomous - plans and executes multi-step workflows without hand-holding
  • Handles both web browsing and code generation in one agent
  • Generous free tier lets you test it on real work before paying
  • Backtracks intelligently when it hits dead ends instead of looping
  • Returns structured data (tables, CSVs) not just prose summaries

Watch out

  • Slow when web scraping is involved - expect minutes per task
  • Lacks domain expertise to evaluate quality, just aggregates surface signals
  • Free tier rate limits kick in quickly with daily use
  • Occasionally misinterprets site structure or fetches stale cached data
  • Broad capability means it's not as sharp as specialist tools for code or research

Use cases

  • Research tasks that need multi-source browsing
  • Drafting reports with live data
  • Competitor intelligence gathering
  • Complex purchase decisions with trade-off analysis