Manus AI

User Guide -- Tips, How-Tos & Best Practices

Zoom

What Manus Is

Manus is a fully autonomous AI agent -- you give it a goal and it works independently in the cloud to complete it. Unlike chatbots, it does not wait for your input at each step. It uses a browser, terminal, code runner, and filesystem to deliver finished outputs asynchronously.

Once you start a task, you can close the app or walk away. Manus continues running in the cloud and notifies you when the result is ready -- returning a finished report, spreadsheet, web app, or file rather than a chat message.

Autonomous agent Runs in cloud Async execution Browser + terminal Delivers files Scheduled tasks MCP integrations Wide Research
⚠ Important: Treat all Manus output as a first draft. Verify facts, sources, and any data before using output in high-stakes contexts. Manus is a powerful starting point, not a finished product.
Data and privacy note: Manus lacks SOC 2 and ISO 27001 certification as of mid-2026. Do not submit confidential PII, regulated data, or sensitive internal documents without reviewing Manus's current data handling policies.

Manus as a Fully Autonomous Agent

Manus is not a chatbot. It is a general-purpose AI agent that receives a goal, plans a sequence of steps, executes them in a cloud sandbox, and returns a finished deliverable -- without requiring your input at each step.

⚙️ How Manus Works

🧱 Multi-Agent Architecture
A Planner agent decomposes your goal into subtasks. Executor agents carry out each subtask in parallel or in sequence using a browser, terminal, filesystem, and code runner -- all inside an isolated cloud VM.
🛠️ Cloud Sandbox
Each task runs in a sandboxed Ubuntu virtual machine. The VM has a full Chromium browser, Python/JavaScript/Bash execution, filesystem access, and networking. Paid user sandbox state persists for up to 14 days.

🕐 Tasks run asynchronously. You can close the app, shut your laptop, and return later. Manus notifies you when work is complete.

Core Capabilities

CapabilityWhat Manus delivers
Web research Autonomously browses live sites, aggregates dozens of sources, produces cited research reports
Wide Research mode Deploys 100+ parallel sub-agents for high-volume tasks: profile 200 companies, compare 100 products
Coding & apps Writes and runs code; builds full-stack web apps (React, Flask, SQLite) with authentication and deployment
Data analysis Accepts CSV, Excel, JSON, PDF; runs Python to produce charts, pivot tables, and summaries
Document creation Generates polished PDF reports, Word docs, Manus Slides (with AI-generated visuals)
Browser automation Fills web forms, navigates multi-step workflows, extracts structured data from web apps
Scheduled tasks Set a recurring job (daily report, weekly digest) and Manus runs it automatically on schedule
Mail Manus Forward an email to your Manus address; it reads the context and attachments and executes the task

Execution Environments

EnvironmentBest for
Temporary Sandbox Standard research, writing, coding, file generation, and web app building
Manus Desktop Tasks needing your local files, local apps, or your local browser session
Cloud Computer 24/7 bots, scheduled scripts, persistent databases, and tools that run even when your machine is off

Operating Modes

ModeDescriptionCredit Cost
Chat Conversational Q&A only, no execution 0 credits
Adaptive Auto-escalates to Agent Mode when complexity warrants Variable
Agent Full autonomous execution in the sandbox Per task
Wide Research Parallel multi-agent for large-scale research (10+ items) Higher

How Manus Differs from ChatGPT, Claude & Perplexity Computer

The autonomy spectrum: from chatbot to autonomous agent

Autonomy Comparison Table

DimensionChatGPT / ClaudePerplexity ComputerManus
Interaction model Reactive: prompt / respond cycle Conversational with tool use Autonomous: goal in, deliverable out
Human involvement Required at every step Moderate: user stays present Minimal: submit and leave
Execution Text generation + limited tool calls Cloud sandbox, browser, code Full Ubuntu VM: browser, terminal, filesystem, networking
Output type Text, code, images Text, files, web results Finished deliverables: reports, apps, spreadsheets, deployed sites
Parallelism None Limited Wide Research: 100+ parallel sub-agents
Scheduling None None Recurring scheduled tasks, 24/7 cloud bots

Key Distinctions

  • vs. ChatGPT: ChatGPT requires manual prompting at each step. Manus handles the entire workflow. For large-scale research (50+ items), ChatGPT quality degrades as context fills; Manus Wide Research gives each item a fresh context and dedicated agent.
  • vs. Claude: Claude's computer use mode requires you to remain present and guide each step. Manus is "autopilot" -- submit the goal and return to completed output. For deep local codebase work, Claude Code is more appropriate.
  • vs. Perplexity Computer: Perplexity Computer is designed for interactive, conversational task assistance. Manus is designed for asynchronous, hands-off execution with explicit parallelism, scheduled tasks, and MCP connectors.

Example Tasks: How-To Cards

Real prompt patterns for common Manus workflows.

01
Competitor research report
Manus browses live sites, extracts pricing and features, and delivers a formatted spreadsheet or PDF.
"Research the top 10 SaaS CRM competitors. For each, extract pricing, key features, target market, and G2 rating. Deliver a formatted Excel spreadsheet."
02
Deep research report with citations
Manus aggregates dozens of sources into a structured, cited PDF report on any topic.
"Write a 15-page research report on [topic], citing at least 20 primary sources. Include an executive summary, key findings, and a references section. Deliver as PDF."
03
Data analysis from a spreadsheet
Upload a CSV or Excel file and get back charts, pivot tables, and a plain-language summary.
"Analyze the attached CSV. Identify the top 3 revenue drivers, flag anomalies, and produce a one-page summary with charts. Deliver as PDF."
04
Build and deploy a web app
Manus writes the code, sets up the backend, and gives you a live URL for a functional prototype.
"Build a React web app with a Flask backend and SQLite database that tracks [use case]. Include user login. Deploy it and give me the live URL."
05
Wide Research: 50-company profile
Wide Research mode deploys parallel agents to research many companies or products simultaneously.
"Research the top 50 fintech startups in Southeast Asia. For each, extract founding year, funding stage, headquarters, and product focus. Deliver as a structured spreadsheet."
06
Slide deck generation
Manus researches the topic, structures the outline, and generates slides with AI visuals.
"Create a 12-slide presentation on [topic] for [audience]. Include a title slide, executive summary, 8 content slides with supporting data, and a next-steps slide."
07
Scheduled weekly digest
Set up a recurring task and Manus delivers a fresh report to your email or Slack every week automatically.
"Every Monday at 8 AM, research the top 5 AI news stories from the past week and email me a 300-word digest with links."
08
Email-triggered task with Mail Manus
Forward an email to your Manus address and it reads the full context and attachments to complete the task.
Forward a client email to your Manus address with the note: "Convert this RFP into a project proposal outline with timelines and deliverables."

Native MCP Connectors

Manus uses the Model Context Protocol (MCP) as its integration standard. Once connected via OAuth, a connector is available in all conversations and Projects -- no copy/paste required.

🔗 Available MCP Connectors

EmailGmail
CalendarGoogle Calendar
File StorageGoogle Drive
Knowledge / DocsNotion
CRMHubSpot
Code / Version ControlGitHub
PaymentsStripe
Team MessagingSlack
AdvertisingMeta Ads Manager
DatabasePostgreSQL

🔒 Connector data is kept within your session. Sensitive output is hidden in shared sessions and API keys are auto-redacted. Full compliance-grade audit trails are not yet available.

Browser Operator Extension

The Browser Operator is a Chrome/Edge extension that turns your local browser into Manus's agent surface. It enables access to tools behind paywalls and authenticated sessions without needing an API integration.

  • Works with SEMrush, Ahrefs, LinkedIn, internal tools, CRMs
  • Uses your existing cookies and sessions -- no re-authentication needed
  • Required for any authenticated SaaS tool without a native MCP connector
  • Available on Pro and Team plans

Zapier (8,000+ Apps)

Connect Manus to over 8,000 apps through Zapier trigger/action workflows. Manus does not require a native connector for every tool -- Zapier bridges the gap.

  • Example: "Create a Manus task when a new message arrives in Slack"
  • Example: "When a new row is added to Google Sheets, trigger a Manus research task"
  • Allows two-way data flow between Manus and your existing tool stack

Manus API (Beta)

Programmatic access for developers building custom applications on top of Manus's autonomous execution engine.

  • API key authentication; documented at open.manus.ai/docs
  • Supports structured output schemas for machine-readable results
  • Task polling and event-based task management
  • Enables building Manus-powered internal tools and automations

Browser-Based Access (No API Key)

Because Manus controls a full browser, it can interact with any web-based application without a dedicated connector -- including tools with no public API.

  • Google Sheets, Google Docs
  • Airtable, Salesforce
  • Any web app Manus can log into or navigate

Projects: Persistent Context Workspaces

Projects bundle instructions, files, connectors, and skills into a persistent workspace. Every task run inside a Project automatically inherits that context.

  • Store your SOPs, reference documents, and connector settings once -- never re-explain context
  • Project Skills lock a specific set of workflows to a project, preventing unrelated workflows from interfering
  • Ideal for recurring work: weekly reports, client deliverables, ongoing research programs
  • Team admins can publish Projects with gold-standard Skills shared across the whole team
Tip: Create one Project per recurring work type (e.g., "Weekly Competitor Report," "Client Onboarding"). Store all context, connectors, and skills there so new runs start with zero setup.

Skills: Reusable Workflow Packages

Skills are reusable agent workflow packages that act as operating manuals. They extend what Manus can do within a domain and can be triggered on demand.

  • Created by saving a successful workflow: "Package this workflow into a Skill"
  • Triggered with a slash command: /SKILL_NAME
  • Can be shared across a team or imported from GitHub
  • Official Manus skill library maintained for common use cases (Essential and Pro skill sets)
  • Skills + MCP Connectors work together: Skills provide the procedural logic, Connectors provide the data pipeline

Skills Use Cases

  • Weekly market research digest (same format every time)
  • Lead enrichment from a list of company names
  • Invoice extraction from forwarded emails
  • Competitive pricing update on a set schedule

Plan Comparison

All plans include Chat Mode (always free, no credits). Credits are consumed only when Manus runs Agent, Adaptive, or Wide Research mode tasks.

Plan Price Monthly Credits Daily Refresh Concurrent Tasks Scheduled Tasks Models
Free $0 0 (+ 1,000 one-time starter) 300/day 1 2 Manus 1.6 Lite only
Pro Entry $20/mo 4,000 300/day 20 20 Lite + 1.6 + 1.6 Max
Pro Mid $40/mo 8,000 300/day 20 20 Lite + 1.6 + 1.6 Max
Pro Top $200/mo 40,000 300/day 20 20 Lite + 1.6 + 1.6 Max
Team $20/seat/mo (min. 2 seats) 8,000+/seat 300/seat 20 (shared) Not stated Lite + 1.6 + 1.6 Max

Annual billing saves approximately 17% on all paid tiers (roughly two months free per year). A 7-day free trial is available on Pro Mid ($40) only.

The Credit System

Credits are the unit of consumption for everything Manus executes. The per-credit rate is approximately $0.005 per credit at all Pro tiers.

What Consumes Credits

  • LLM tokens -- task planning, decision-making, output generation
  • VM compute -- cloud workspace runtime (browser automation, code execution, file operations)
  • Third-party API calls -- integrated services like Similarweb, financial data providers

Estimated Credit Costs

Task TypeApproximate Credits
Chat query0 credits
Simple research summary50 - 100
Blog article draft100 - 200
Deep report (10+ sources)150 - 400
Slide deck generation200 - 500
Complex multi-step workflow300 - 900
Full-stack web app build500 - 2,000
Wide Research (50+ items)500 - 2,000+

Credit Rules to Know

Monthly credits do not roll over
Unused monthly credits expire at the end of each billing cycle. Plan your usage accordingly -- save large tasks for later in the month only if you have credits to burn.
Daily refresh credits reset every 24 hours
The 300 daily refresh credits do not accumulate day to day. They reset each day whether or not you used them. The one-time 1,000 starter pack never expires while your account is active.
Failed tasks still consume credits
Credits are charged even if a task fails -- including failures caused by Manus infrastructure problems. Manus has offered partial (50%) refunds for server-side failures rather than full credit restoration.
No hard budget cap
Manus shows an estimated credit range before launching a task, but there is no "stop at X credits" mechanism. Actual costs can exceed estimates. Monitor large tasks carefully.

Plan Selection Guide

If you...Choose
Want to test ManusFree (1,000 starter + 300/day)
Run 1-2 deep research tasks per weekPro Entry ($20/mo)
Run daily autonomous workflowsPro Mid ($40/mo)
Run constant research or near-production loadsPro Top ($200/mo)
Have a team of 2+ needing shared workflowsTeam ($20/seat/mo)

Team Plan Features

FeatureDetails
Price$20/seat/month (min. 2 seats); annual billing saves 17%
Monthly credits8,000+ per seat
SSOIncluded
Access controlsIncluded
Shared credit poolsAny team member can draw from the shared pool
Usage analyticsAdmin visibility into who is consuming credits and on what
Shared Skills libraryTeam admins can publish Skills to all members
Central billingSingle invoice for the team
Wide Research, Slides, deploymentIncluded

What the Team Plan Does NOT Include

  • Missing SOC 2 certification
  • Missing ISO 27001 certification
  • Missing On-premises deployment
  • Missing GDPR data processing agreements (not publicly documented)
  • Missing Dedicated infrastructure (shared with consumer users)
  • Missing SLA guarantees
⚠ Billing caution: Canceling your account via account deletion does NOT cancel your subscription billing. Cancel your subscription separately before deleting your account to avoid continued charges.

Chat Mode is always free and fully functional even when your monthly credit pool is exhausted. Agent Mode tasks require available credits.

The 5-Part Prompt Formula

Manus is an agent, not a chatbot. Prompts that work for ChatGPT often underperform with Manus because they do not specify the deliverable clearly enough. Use this five-part formula to get consistently strong results.

🏃 The Five-Part Task Formula

1. Role / Context
"You are a market research analyst with expertise in SaaS..."
2. Clear Goal
What the finished output should be, not what steps to take.
3. Background / Scope
Constraints, data sources, time period, geographic focus.
4. Output Format
"Deliver a PDF report," "create a spreadsheet with columns X, Y, Z," "build a site and give me the URL."
5. Authorization Scope
What Manus is and is not allowed to access or do during this task.

Prompt Improvement Examples

Weak PromptStronger Prompt
"Research competitors" "Research the top 10 SaaS CRM competitors. For each, extract pricing, key features, target market, and G2 rating. Deliver a formatted Excel spreadsheet."
"Write some blog posts" "Write 5 blog post outlines on [topic] targeting [audience]. Each outline should have a headline, 3-sentence summary, and 5 section headers. Deliver as a Google Doc."
"Analyze this data" "Analyze the attached CSV. Identify the top 3 revenue drivers, flag any anomalies, and produce a one-page summary with charts. Deliver as PDF."

Good Habits for Better Results

Use Chat Mode before committing credits
Chat Mode costs zero credits. Use it to refine your task scope, ask questions, and plan before committing to Agent Mode execution. A few chat turns can prevent a costly failed run.
Break large projects into stages
Run Stage 1, review the output, then pass results into Stage 2. This lets you catch errors early without burning credits on a derailed workflow. Very large multi-day tasks have higher failure rates.
Use Projects for recurring context
Store your SOPs, reference documents, and connector settings in a Project. Every task run inside it automatically inherits that context -- no re-explaining needed.
Use Skills for repeatable workflows
If you run the same type of analysis weekly, package it as a Skill and trigger it with a slash command. Skills + Connectors deliver consistent, zero-setup recurring outputs.
Monitor early runs before going hands-off
Watch the live execution window ("Manus's Computer") for the first few runs of a new task type. Once you trust the pattern, submit and walk away.
Treat all output as a first draft
Manus is excellent at research and structure, but final outputs benefit from human review -- especially for accuracy, tone, and domain-specific nuance. Always verify key claims before using output externally.

Known Limitations

  • CAPTCHA Manus fails gracefully on sites with aggressive anti-bot protection. The Browser Operator partially mitigates this for sites where you already have a local authenticated session.
  • App quality Manus-built web apps are functional proofs of concept, not production-ready products. Plan for engineering review before deployment.
  • No SOC 2 Manus lacks SOC 2 and ISO 27001 certification as of mid-2026. Not suitable for regulated data or highly sensitive internal documents.
  • Inconsistent quality Task results can vary meaningfully between runs, even with identical prompts. Credit cost for comparable tasks has been documented varying from 180 to 340 credits for the same task type.
  • Slow under load Complex tasks take 10-30+ minutes. Server load during peak periods slows execution further and can cause task failures or sandbox freezes.
  • No budget cap There is no mechanism to stop a task at a credit limit. Actual costs can exceed pre-task estimates.
  • No inbound webhooks Manus cannot react to external events in real time. Tasks must be initiated manually or via scheduled recurring runs.

Autonomy Level Tip Cards

Match the right tool to the right task.

🟢 Use Manus when...
The task has a clear deliverable, runs over a long time horizon, requires browsing many sources, needs to produce files, or should repeat on a schedule. You want to walk away and return to a result.
🟡 Use Chat Mode (free) when...
You want to explore ideas, refine a prompt, ask quick questions, or plan a workflow before committing credits. Chat Mode always costs zero and has no execution.
🔴 Do NOT use Manus when...
The task requires real-time customer interaction, financial transactions, submission of regulated PII, deep local codebase integration, or any workflow where CAPTCHA-protected sites are the primary data source.
For deep local codebase work: Claude Code or Cursor is more appropriate than Manus. Manus excels at general autonomous workflows that extend beyond code -- research, data, files, and multi-service integrations.
🎓 Manus AI Resources

Official documentation, tutorials, and community resources for getting started with and mastering Manus AI.

Manus AI – Official Docs 📄 Documentation · Free Manus Documentation: Welcome & Introduction Official Manus documentation covering what Manus is, how the autonomous agent works, and where to start. Includes the full documentation index for deeper exploration. Read Docs → Manus AI – Official Site 🏠 Product Home · Free Tier Available Manus AI Official Site The official Manus product home. Access Chat Mode (always free), review capability examples, and start your first autonomous agent task directly from the browser. Visit Manus → YouTube · May 2026 📹 Video · 13 min · Beginner Manus AI Explained Simply in 13 Minutes: Setup & Use Cases Clear 13-minute beginner walkthrough of Manus AI: setup, real use cases, how the autonomous agent handles multi-step tasks, and what sets it apart from standard chatbots. Watch on YouTube → YouTube · Dec 2025 📹 Video · Complete Guide · Intermediate Manus AI Tutorial 2025: Complete Guide & Real Use Cases Comprehensive guide testing Manus AI on real tasks: software building, web browsing automation, and task execution. Covers setup, limitations, and practical use cases for developers and beginners. Watch on YouTube → YouTube – Official Manus Channel 📹 Official Channel · Free Manus AI Official YouTube Channel The official Manus AI YouTube channel featuring product demos, capability showcases, and tutorials published directly by the Manus team. View Channel → Manus AI – Official Docs 📄 Feature Guide · Free Manus Website Builder: Getting Started Official Manus documentation for the Website Builder feature: how to get started building web projects using Manus AI's autonomous capabilities. Read Guide →

Resources are provided for informational purposes. Always verify content is current before sharing.

This guide is for training and orientation only. Features, limits, pricing, and integrations may change as Manus updates its platform. Always review Manus's current terms, data handling policies, and acceptable use guidelines before submitting sensitive data. Chat Mode is always free. Content as of May 2026.