Manus is an AI agent with documented task, project, research, browser, desktop, data and presentation surfaces. Check the current account and pricing page before relying on a plan, credit allowance, connected account or local-folder access.
Used this tool? Rate it
Used this tool? Rate it
Current official documentation describes Manus as an autonomous general AI agent intended to complete tasks and deliver results. That makes the Manus AI agent a different product shape from a chat-only interface: the documentation describes a virtual colleague that can plan, execute and deliver work products. Availability still depends on the current product, account and route; that positioning does not promise that every feature is enabled for every user.
The current documentation describes the core runtime as a sandbox virtual computer with internet access, persistent files and installable software or custom tools. That runtime description is useful when choosing the right unit of work: a task can have files, browser steps and a result, rather than only a prose answer. It does not by itself establish additional security, retention or policy terms. For those questions, use the current product controls and the applicable policy.
Manus has several documented product surfaces. Projects keep recurring task context; Skills package reusable instructions and resources; Slides, data analysis and multimedia target particular deliverables; Scheduled Tasks run at stated times; Cloud Browser and Desktop extend where work can happen. The useful evaluation question is therefore not whether an agent sounds autonomous, but which documented surface matches the intended input, output and permission boundary.
For AI task execution, begin with an output that maps to one current surface. A CSV analysis can be scoped as an analysis request with a dashboard or report output. A presentation request can be scoped around a supplied brief and a PPTX export. A recurring digest belongs to Scheduled Tasks only after its timing and output are specified. These are product-specific paths rather than a claim that the agent is suitable for every decision or action.
The public documentation separates task workspaces, reusable Skills, research, data analysis, slides, multimedia, scheduled work, cloud browsing and Desktop. The pricing Help Center describes Free, Pro and Team while directing commercial conditions to the displayed pricing page. API v2 is a separate developer surface. Check the relevant current route and account state for any plan, connector, account or permission condition.
Projects are persistent workspaces with shared instructions and files, and each new task inherits that configuration. The current documentation presents a Project as a reusable setup: create a name, write a master instruction, add a knowledge base of files and documents, then create tasks inside it. This is the product basis for a recurring autonomous agent workflow when the same context should be applied to new tasks without entering it again each time.
Project configuration has a documented timing distinction. Instruction updates apply the next time a message is sent in a current task. File updates take effect only for new tasks created after the update, while earlier tasks keep the configuration that existed when they were created. That distinction matters when a task depends on a specific reference-file version: an editor should not assume a changed upload silently rewrites an earlier run.
Projects and individual tasks are private by default according to the current Projects page. When a person is invited to a Project, the documentation says they receive the shared master instruction and knowledge base but see only tasks they created unless a specific task is separately shared. Treat this as a current feature statement with the account interface as the final source for a particular invitation, rather than a broad legal privacy claim.
For Manus task planning, create separate Projects for genuinely different recurring inputs instead of placing unrelated files under one master instruction. The current official Projects announcement gives weekly reports, content pipelines, competitive analysis and prototypes as examples of recurring work. The product-specific benefit is context inheritance and task filtering; it is not a promise that the content of an uploaded file is correct, complete or appropriate for every new task.
Current Skills documentation describes modular, file-system-based resources. It says a Skill can be made from a successful interaction, uploaded as a zip file, .skill file or folder, taken from the official library, or imported from GitHub. A Skill is therefore an explicit, inspectable resource in the documented product—not evidence that its instructions, code or shell commands are safe to run without reading them.
The current Skills documentation says community Skills should have their contents verified before use because they can contain code and shell commands. For an evaluation, record the source of a Skill, inspect its SKILL.md and associated files, and test it with permitted material. That advice is tied to the documented Skills surface, not a generic claim about every third-party integration.
Wide Research is documented as parallel processing for many similar items, with dedicated context for each independently processed item. The current description uses examples such as analysing many products, researching companies or generating content across comparable items. It is a different path from asking one agent to process a long list sequentially: the product description says independent agents work in parallel and each receives its own context.
This makes Manus an AI research workspace when the work can be represented as repeated records. For example, a request can name a fixed set of companies, define fields to collect, request one row per company, and ask for a separate flag for an unavailable source. The documented parallel architecture does not validate the resulting research; readers should still inspect important citations and confirm what was actually available to the task.
For data work, the current documentation describes uploading a file, stating what to learn, and receiving a slide, report, dashboard or webpage output. It names CSV and Excel among the data-file examples and offers prompts such as identifying growth opportunities, comparing regions or finding patterns in customer behaviour. This is a product-specific workflow: file → stated analytical question → output format. It does not mean every file is accepted, every pattern is reliable, or a chart is a substitute for checking the underlying data.
The documented Data Analysis & Visualization surface lists presentation-ready visual insights and output formats including slides, detailed reports, interactive dashboards and standalone webpages. When assessing it, choose one output and one question first: a comparison of named regions, a trend over a stated period, or a segmentation using columns that are already present in the file. The account interface is the place to confirm accepted format, capacity and any current task limit.
The current Wide Research page describes a system for large-scale research and processing tasks, not a guarantee of continuous monitoring. Use it for a bounded comparable set whose count, fields and completion condition can be named. A request such as “analyse the supplied 50 entries with these five fields” matches the documented one-item-per-agent framing more directly than an open-ended request to keep watching the web indefinitely.
The Slides documentation lists research, content, design, visuals and speaker notes, with PPTX, PDF, web slides and a speaker-notes document among the listed exports. It also documents importing a PowerPoint template (.pptx) so that Manus applies the template’s layouts, colours and fonts. This supports a concrete slide workflow: state the audience and objective, provide a permitted template if one is required, inspect the generated deck, and export the format that will actually be edited or delivered.
The current Slides page lists PowerPoint (.pptx) as fully editable in PowerPoint or Google Slides. That is a useful handoff property for a presentation team, but it is not a claim that an automatically generated deck is factually complete or presentation-ready for every context. Check citations, charts, numbers, speaker notes, required brand elements and the final rendered file before it is used.
Current Multimedia Processing documentation lists image generation, image understanding, video understanding, voice output and speech-to-text. It describes a multi-mode workflow that can combine text, images, video and audio. The feature list supports selecting a media task—for example, transcription of a permitted recording or extracting information from an image—not asserting rights to upload a particular asset or use a generated person, voice or visual in a particular jurisdiction.
The current Slides page explicitly recommends being specific about audience, goal, length and timing. A short brief can therefore say who will see the deck, what decision or explanation it should support, which source materials are permitted, the requested language, and whether the deliverable is PPTX, PDF, web slides or notes. Those details use the documented Slides surface rather than adding an invented presentation-management feature.
For multimedia, name the operation rather than only saying “work on this media”: generate an illustration, understand an image, examine a video, create voice output or transcribe speech. The official table distinguishes these operations. Confirm the current upload interface, output availability and account conditions before submitting material; the documentation’s feature list does not supply a universal usage-rights answer.
Scheduled Tasks run at stated times rather than as continuous real-time monitoring, and the settings document pause, edit, delete and execution-history controls. The current page gives daily, weekday, weekly, monthly, custom and one-time examples. It also says that a schedule can send results by email, post to Slack, save to Google Drive, update a spreadsheet or send via a connector; those destinations must be checked in the current account before relying on them.
The documented boundary is important: recurring research, regular reports, periodic data collection and change checks are listed uses, while real-time monitoring is explicitly excluded because scheduled runs occur at specific times. A schedule should therefore include a cadence, time zone, source window, output route and a way to inspect the previous run. It should not be described as a live alerting system merely because it repeats.
The current interface instructions say active schedules can be viewed under Settings → Scheduled Tasks. The same page documents toggling a schedule off without deleting it, changing its timing, task description or output method, deleting it, and viewing prior runs, results and errors. These product controls provide useful lifecycle points for a recurring task; confirm that the relevant control is available to the intended account before making it part of a workflow.
Cloud Browser is documented as a cloud environment that can visit pages, click, fill forms and extract data; use of authenticated accounts is an authorization decision. Its page says a user can log into personal or work accounts in the browser to authorize tasks requiring authentication. That makes browser automation a permissioned product surface, not a reason to assume the agent may safely act on every account or submit every form.
The current Cloud Browser documentation says it operates from data center IP addresses, not residential IPs, and distinguishes it from My Browser for some sites. A task can fail or appear differently because a site treats a data-center connection differently. Before depending on a route, confirm the site’s own rules, sign in only when the current product asks for authorization, and observe the first run. These are direct implications of the documented Cloud Browser surface rather than claims about a specific third-party site.
On Desktop, My Computer is documented to work with a selected local folder only after the operating system asks the user to confirm access. The Desktop documentation says the feature can interact with local files, applications and development tools, and its getting-started sequence is download, sign in, select a folder, then approve the operating-system prompt. The scope is therefore the folder selected by the user; do not describe Desktop as blanket access to a whole machine.
The current Browser Operator documentation says it works directly within the current browser session, using existing logins and active tabs. Cloud Browser and Browser Operator are separate documented routes. A person evaluating an authenticated task should identify which route is visible in the account, what account or folder is being shared, the exact action requested, and whether the action can be stopped before an external submission occurs.
Current Desktop documentation says the app is available for macOS and Windows, but platform availability and account eligibility must still be confirmed at the time of use. The current Desktop credit Help Center says the app uses the same credit system as the web platform and mirrors the web-dashboard balance without a separate fixed credit bundle. Treat this as a current product condition, not as a reason to infer local execution is free.
The current Help Center lists Free, Pro and Team options, while the displayed pricing page governs the commercial terms. For a personal trial, the current Free description lists Chat Mode, Manus 1.6 Lite in Agent Mode, 300 daily refresh credits, one concurrent task and two scheduled tasks. The page is dated 16 March 2026 and gives Free as $0/month, but eligibility, availability and checkout details must be checked in the live account and price page.
The same current article lists two Pro starting-price descriptions: from $20/month with an annual-billing discount and from $40/month with a seven-day free trial and annual-billing discount. It lists Manus 1.6 Max, Manus 1.6 and Manus 1.6 Lite in Agent Mode, monthly credits starting at 4,000 or 8,000 respectively, Advanced Research, website deployment, slide generation, Wide Research, and 20 concurrent and 20 scheduled tasks. These are current Help Center statements, not a claim that every market, existing subscriber or account sees identical terms.
Manus says credits depend on task complexity and duration and can reflect LLM tokens, virtual machines and third-party APIs during active processing. The current Credits Help Center says completed tasks and the storage or deployment of their outputs do not consume credits. Check the account balance before a substantial task rather than treating a general product description as a cost quote.
The current Credits Help Center says credits are consumed in the order event, daily, monthly, add-on, then free; monthly credits refresh on the subscription date, Free credits never expire, and event credits expire when the event ends. It says add-on credits can carry forward only while an active paid subscription continues. Check the balance and current rules in the account before starting a costly or timed task because plan displays and task complexity can change.
Before a run, define the intended output, inspect the account’s displayed balance and conditions, and keep the task scope bounded. Credit consumption depends on the documented processing inputs and should not be treated as a fixed quote for a different task.
The current API v2 documentation describes programmatic task creation, follow-up messages, result retrieval, file attachments, webhooks, Skills and Agents. The API is a separate developer route. For an API project, use the current API documentation for authentication, endpoint, rate-limit, billing and data-handling questions rather than applying an app-plan statement to programmatic use.
The official documentation index currently lists Browser Operator, Slack Integration and Zapier routes. A route in that index establishes that a document exists, not that a particular connector is enabled for a given account or has a particular permission scope. This is why the Slack-integration and API tags remain discovery labels with a bounded interpretation.
The documented agent, sandbox, browser and Desktop surfaces can create files, handle data, or act through an authorized account. None of those feature descriptions establish that a result is accurate, that a third-party site permits the action, or that a current account has access. Use the current service interface for availability, account controls and consent flows; use the material owner’s rules for a file or account; and use an official current policy for any legal or data-rights question that the feature page does not answer.
Feature pages should be read within their stated surface. Plan details do not create identical rights for every account, and API documentation does not establish web-app billing or data terms. For a material permission, data-rights or service-policy question, consult the applicable current product control or official policy.
The current official documentation calls Manus an autonomous general AI agent for completing tasks and delivering results. Its public documentation then separates concrete surfaces such as Projects, Slides, data analysis, schedules, browsers and Desktop. Check which surface is actually available in your account.
It establishes the documented runtime description: internet access, persistent files and installable software or custom tools in a virtual computer. It does not establish additional legal, security or policy terms beyond what the applicable policy and account controls say.
Projects provide shared instructions and files. The current documentation says new tasks inherit the Project configuration; instruction edits apply when the current task next receives a message, while file edits affect new tasks created afterwards.
Wide Research is documented for large comparable sets and describes independent agents processing individual items in parallel. Define a bounded list and a repeatable output schema; then inspect the results and sources that matter to the decision.
The current documentation describes uploading data, stating what to learn, and selecting slides, reports, dashboards or webpages. Confirm accepted file details and current account limits in the live product.
The Slides page lists editable PPTX, PDF, web slides and a speaker-notes document. It also documents uploading a PPTX template. Review the generated material and exported file before presenting it.
No. The current Scheduled Tasks document says runs occur at stated times rather than continuously. It documents daily through custom and one-time schedules as well as pause, edit, delete and history controls.
The documentation says a user can log into personal or work accounts in Cloud Browser and thereby authorize Manus to perform authenticated tasks. Treat the sign-in and the requested action as an explicit permission decision, not as a default capability.
My Computer is documented to access a local directory selected by the user after an operating-system confirmation prompt. It is not documented here as unrestricted access to every local file.
Current Help Center pages list plan details and say credits depend on task complexity, duration and active processing resources. Check the live account and displayed pricing for the actual plan, balance, renewal, eligibility and task conditions before buying or running work.
