← Chat

ManusManus

Manus AI（Monica 团队）Manus AI (Monica team)

订阅+积分Subscription + credits 云端沙箱Cloud sandbox 异步任务Async tasks 通用agentGeneral-purpose agent 产物发布Artifact publishing partial lifecycle: active 核验Checked 2026-10-22

Manus 是「整段任务扔给云端沙箱」的通用任务 agent：云 VM 里执行代码/浏览/产出/发布（manus.space），模型由产品侧编排不开放选择；积分制，异步是本体形态。

Manus is the general-task agent you hand whole tasks to in a cloud sandbox: code execution/browsing/output/publishing (manus.space) run in a cloud VM, models are orchestrated by the product with no user selection; credits-based, async is its core form.

适合与不适合When to use it, when not to

适合：研究报告、数据整理、建站发布类整段任务异步外包。不适合：要选模型、要本机文件参与、要免费大量用。

Fits: whole-task outsourcing (research reports, data wrangling, site building/publishing) to a cloud agent asynchronously. Not a fit: model selection, local files in execution, heavy free use (credits).

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-22

出品属地Origin: 国产domestic　Manus AI（Monica 团队；主体沿革见官方 Terms，本站未逐条核验注册地细节）。Manus AI (Monica team; entity details per official Terms, registration specifics not verified line by line).
国内使用Direct access: 大陆可直连direct OK　manus.im 大陆直连实测 200（2026-10-22）。manus.im measured 200 direct from mainland (2026-10-22).
备案状态CAC filing: 备案未核验filing unknown：未核验到 Manus 在网信办公示名单的条目（2026-10-22 检索）；其面向大陆用户的服务形态与备案状态官方未声明——「未核验到」不等于「未备案」。no Manus entry verified in the CAC public registry (searched 2026-10-22); its mainland-facing service form and filing status are not declared by the vendor — "not verified" does not mean "not filed".

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 模型由 Manus 侧编排路由（官方未公开完整模型清单）；用户不直接选模型—— 「用什么模型干活」是产品内部决策，这是它与通用对话产品的关键差异之一。
Models are orchestrated and routed on the Manus side (no complete model list published officially); users do not pick models directly — "which model does the work" is an internal product decision, one of its key differences from general chat products.
运行位置Where it runs: 任务跑在 Manus 的云端沙箱（manus.computer 域名的 sandbox 面直抓所见）： agent 在云 VM 里执行代码、浏览网页、产出文件；网页端下达与跟进，本机不装东西。
Tasks run in Manus's cloud sandbox (the sandbox surface on the manus.computer domain, as seen via direct fetch): the agent executes code, browses the web and produces files inside a cloud VM; the web UI issues and follows up, nothing is installed locally.
本地文件Local files: 产物（文件/网页/报告）生成在云端沙箱，可下载导出；上传数据存于账户侧。与「本机 agent」完全相反——执行面在厂商云上。
Artifacts (files/web pages/reports) are generated in the cloud sandbox and can be downloaded/exported; uploaded data is stored on the account side. The exact opposite of a "local agent" — the execution surface is on the vendor's cloud.
关机后的任务Tasks after shutdown: 异步任务是它的本体形态：任务提交后可离开页面，agent 在云端继续跑（官方叙事「把任务交给它」）；有宽限期/排队机制的产品化痕迹（积分制与并发限制）。
Asynchronous tasks are its native form: after submitting a task you can leave the page and the agent keeps running in the cloud (the official narrative is "hand the task to it"); there are productization traces of grace-period/queueing mechanics (credit system and concurrency limits).
工具与扩展Tools & extensions: 自带浏览器、代码执行、文件操作、部署（manus.space 发布面）等工具面；连接外部服务（邮件/日历等）经官方集成线。MCP 支持未逐条核验。
Built-in tool surfaces incl. browser, code execution, file operations and deployment (the manus.space publishing surface); connecting external services (email/calendar etc.) goes through official integration lines. MCP support not verified item by item.
上下文与记忆Context & memory: 以任务（session）为单位组织工作；跨任务记忆机制未核验到官方声明。
Work is organized per task (session); no official statement verified for a cross-task memory mechanism.
权限与限制Permissions & limits: 账号+订阅体系；云端沙箱有联网能力（浏览/下载是产品语义）；数据留存与训练口径按官方 Terms/Privacy 管辖，细节未逐条核验。
Account + subscription system; the cloud sandbox has network egress (browsing/downloading is product semantics); data retention and training terms are governed by the official Terms/Privacy, details not verified item by item.
适合什么任务Best for: 适合：把「整段任务」（研究报告、数据整理、建站发布）异步交给云端 agent、要产物直接发布（manus.space）。不适合：要选模型、要本机文件参与执行、要免费大量使用（积分制）。
Suits: handing whole tasks (research reports, data tidying, site building & publishing) asynchronously to a cloud agent, with artifacts published directly (manus.space). Does not suit: picking models, local files participating in execution, or heavy free use (credit system).

头号误解Top misconception

把 Manus 当「会聊天的机器人」——官方定位是通用任务 agent（聊天只是入口之一），执行在云端沙箱Treating Manus as 'a chatting bot' — the official positioning is a general-task agent (chat is just one entry); execution happens in a cloud sandbox
与 coding 大类的云异步 agent（Devin/Jules）区分：Manus 主轴是通用任务（报告/建站/数据），不是代码库Distinguish it from coding-category cloud agents (Devin/Jules): Manus's main axis is general tasks (reports/sites/data), not codebases
云端沙箱有联网出口——上传敏感数据前想清楚它会带着这些数据在云上浏览（官方 Privacy 管辖，未逐条核验）The cloud sandbox has a live network egress — think through what data it will carry while browsing before uploading anything sensitive (per official Privacy, not verified line by line)

未知项清单Known unknowns

模型编排清单（官方未公开）
数据留存与训练口径细节（Terms/Privacy 直抓但条款未逐条核验）
跨任务记忆机制（未核验到官方声明）
备案状态（未核验到公示条目）

The model-orchestration list (not published officially)
Details of data retention and training terms (Terms/Privacy direct-fetched but clauses not verified line by line)
The cross-task memory mechanism (no official statement verified)
Filing status (no public-registry entry verified)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-22，内容更新日content updated on 2026-10-22。

类型Type	名称Name	链接Link
official	Manus · 官网（产品定位与能力叙事）Manus · official site (product positioning and capability narrative)	https://manus.im
changelog	Manus · Blog（产品更新线，changelog）Manus · Blog (product update line, changelog)	https://manus.im/blog
official	Manus · Terms（官方条款，2026-10-22 直抓核验）Manus · Terms (official terms, direct-fetch verified 2026-10-22)	https://manus.im/terms
official	Manus · Privacy（官方隐私页）Manus · Privacy (official privacy page)	https://manus.im/privacy

实测记录Measured runs

本档未跑对话任务集；能力结论来自官方文档，不评价回答质量。

实测建议（三条，都指向官方可读页没给的东西）：

任务失败/中断时的积分扣减规则（积分制产品，失败计费是用户最关心的）
云端沙箱的网络出口范围（能访问哪些域、能否访问内网型服务）
产物在 manus.space 上的保留与访问控制

This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to chat products).

Suggested measurements (three, all pointing at things the official readable pages do not give):

Credit-deduction rules when a task fails/is interrupted (a credit-based product; billing on failure is what users care about most)
The network egress scope of the cloud sandbox (which domains it can reach, whether intranet-type services are reachable)
Retention and access control of artifacts on manus.space