MetaGPTMetaGPT
MetaGPT 把「软件公司」写进多 agent 协作（MIT）：产品经理→架构师→工程师→QA 的角色流水线， SOP 编码进 agent 交互。学术热度大（71,097★）；但仓侧主线推送停在 2026-01、release 停在 2025-03—— 团队重心在组织内其他项目上移，生产选型按仓侧现状判断。
MetaGPT writes a "software company" into multi-agent collaboration (PM → architect → engineer → QA, SOPs encoded into role interactions). Huge paper heat (71,097 stars); but repo pushes stopped at 2026-01 and releases at 2025-03 — judge production selections by repo-side facts.
适合与不适合When to use it, when not to
适合：多智能体协作研究、SOP 流水线实验、论文线复现。 不适合：要活跃维护的生产选型、要精细单 agent 控制。
Fits: multi-agent research, SOP pipeline experiments, paper-line reproduction. Not a fit: production selections needing active maintenance, fine-grained single-agent control.
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 2026-10-31
- 出品属地Origin
- 国产domestic 北京深度赋智（DeepWisdom）团队线开源项目（GitHub org FoundationAgents；文档站挂 deepwisdom.ai）。Open-source project of Beijing DeepWisdom (GitHub org FoundationAgents; docs site on deepwisdom.ai).
- 国内使用Direct access
- 大陆可直连direct OK docs.deepwisdom.ai 直连实测 200（2026-10-31）；GitHub 仓大陆直连波动（同日代理可达）。docs.deepwisdom.ai measured 200 direct (2026-10-31); the GitHub repo's mainland reachability fluctuates (200 via proxy the same day).
- 备案状态CAC filing
- 备案未核验filing unknown：框架本体是开源库不面向公众提供服务，备案不适用；其关联产品（DeepWisdom 系应用）的备案状态属另一主体——本站未逐条核验，不下断言。the framework itself is an open-source library not serving the public, so filing does not apply; filing status of associated DeepWisdom products belongs to another entity — not individually verified, no assertion made.
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
多模型接入（各家 API 与本地线，未逐条清点）；它的叙事不在模型， 在「把软件公司流程写进多 agent 协作」（官方仓描述「The Multi-Agent Framework: First AI Software Company」）。
Multi-model access (each vendor's API and local lines, not itemized); its narrative is not the model but "writing a software company's process into multi-agent collaboration" (official repo description: "The Multi-Agent Framework: First AI Software Company").
- 运行位置Where it runs
Python 多智能体框架：按「产品经理→架构师→工程师→QA」的角色流水线协作产出软件工件。 GitHub FoundationAgents/MetaGPT 71,097★（API 核验）；最近推送 2026-01-21、最新 release v0.8.2（2025-03-09）—— 仓侧主线更新明显放缓（组织侧其他仓仍活跃，见 pitfalls）。
A Python multi-agent framework: a role pipeline of "product manager → architect → engineer → QA" collaborates to produce software artifacts. GitHub FoundationAgents/MetaGPT 71,097 stars (API verified); last push 2026-01-21, latest release v0.8.2 (2025-03-09) — mainline repo updates clearly slowed (other org repos remain active, see pitfalls).
- 本地文件Local files
本地跑：pip 安装后在你自己的机器上执行，生成的代码、文档与临时产物都落在你的工作目录； 没有「云端替你跑」的托管执行线（官方未提供托管面）。
Runs locally: install via pip and it executes on your own machine; generated code, documents and temp artifacts all land in your working directory; there is no cloud-runs-it-for-you hosted execution line (the official docs offer no hosted surface).
- 关机后的任务Tasks after shutdown
批式流水线为主（一次任务跑完一条软件生产流程），不是常驻接消息的形态； 研究线活跃（组织仓 CEO-Arena 等 2026-10 仍在推送）。
Batch-style pipelines are the main mode (one task runs through one software-production pipeline), not a resident message-receiving form; the research line is active (org repos such as CEO-Arena still pushing in 2026-10).
- 工具与扩展Tools & extensions
角色动作/代码执行/文档产出工具线内置；MCP 支持状态未逐条核验。
Built-in role-action / code-execution / document-output tool lines; MCP support status not itemized.
- 上下文与记忆Context & memory
SOP（标准作业流程）编码进角色交互是它的核心设计——上下文按角色间消息传递组织（官方文档口径）。
Encoding SOPs (standard operating procedures) into role interactions is its core design — context is organized as inter-role messaging (per the official docs).
- 权限与限制Permissions & limits
代码执行发生在你机器上；多 agent 自动流水线的中间产物要人审再落地—— 全自动「软件公司」的产出质量与责任面在用户侧。
Code execution happens on your machine; intermediate artifacts of a fully automatic multi-agent pipeline should be human-reviewed before landing — the output quality and responsibility surface of a fully automatic "software company" sits with the user.
- 适合什么任务Best for
适合：多智能体协作研究、SOP 流水线实验、软件生成流程的学术复现（论文引用量大）。 不适合：要活跃维护的生产选型（推送/发版放缓）、要精细单 agent 控制。
Fits: multi-agent collaboration research, SOP pipeline experiments, academic reproduction of the software-generation process (heavily cited paper line). Does not fit: production selections needing active maintenance (pushes/releases slowed), fine-grained single-agent control.
头号误解Top misconception
- 活跃度两口径：MetaGPT 仓最近推送 2026-01-21、release 停在 v0.8.2（2025-03）——但 FoundationAgents 组织侧其他仓（CEO-Arena 2026-10-29、OpenManus 58,830★）仍活跃：团队重心在移，选型按仓侧现状写Two activity readings: the MetaGPT repo's last push is 2026-01-21 and releases stop at v0.8.2 (2025-03) — yet other FoundationAgents org repos (CEO-Arena pushed 2026-10-29, OpenManus 58,830 stars) stay active: the team's center of gravity is moving; write selections by repo-side facts
- 星数高≠维护活跃：71,097★ 是论文热度存量——生产选型看 pushed_at 与 releases，不看星数High stars ≠ active maintenance: 71,097 stars is paper-heat inventory — for production selections read pushed_at and releases, not stars
- 组织更名线：GitHub org 从 geekan/MetaGPT 迁到 FoundationAgents（深度赋智团队线）——旧链接会跳Org migration: the GitHub org moved from geekan/MetaGPT to FoundationAgents (the DeepWisdom team line) — old links redirect
未知项清单Known unknowns
- MCP 支持状态（未逐条核验）
- 主线是否转入「仅研究维护」状态（官方未声明；按推送/发版事实记 maintenance，不下断言）
- DeepWisdom 商业产品线与框架的关系（未核验）
- MCP support status (not itemized)
- Whether the mainline has moved into a "research-maintenance only" state (not stated officially; recorded as maintenance per push/release facts, no assertion made)
- The relationship between DeepWisdom's commercial product line and the framework (not 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 2026-10-31，内容更新日content updated on 2026-10-31。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| repo | FoundationAgents/MetaGPT · 仓库（MIT；71,097★，最近推送 2026-01-21，GitHub API 核验）FoundationAgents/MetaGPT · repository (MIT; 71,097 stars, last push 2026-01-21, GitHub API verified) | https://github.com/FoundationAgents/MetaGPT |
| docs | MetaGPT · 官方文档（docs.deepwisdom.ai，「The Multi-Agent Framework」）MetaGPT · official docs (docs.deepwisdom.ai, "The Multi-Agent Framework") | https://docs.deepwisdom.ai/ |
| changelog | FoundationAgents/MetaGPT · Releases（v0.8.2，2025-03-09）FoundationAgents/MetaGPT · Releases (v0.8.2, 2025-03-09) | https://github.com/FoundationAgents/MetaGPT/releases |
| repo | FoundationAgents · 组织页（OpenManus/CEO-Arena 等关联仓活跃度对照）FoundationAgents · organization page (activity contrast with related repos OpenManus / CEO-Arena) | https://github.com/orgs/FoundationAgents/repositories |
实测记录Measured runs
本档未跑集成与性能实测；结论只覆盖 API 与官方文档披露的能力面。
This site has not completed measured runs. The test protocol is at protocols/framework/_protocol.md.
相关条目Related entries
- CrewAICrewAI同大类对照：同为多角色协作框架，CrewAI 面向生产编排且活跃发版，MetaGPT 偏研究流水线且仓侧放缓——活跃度是选型差。Same-category contrast: both are multi-role collaboration frameworks — CrewAI targets production orchestration and releases actively, MetaGPT leans to research pipelines and its repo has slowed; activity is the selection gap.
- AgentScopeAgentScope同大类对照：同为国产多智能体框架，AgentScope（阿里系）仍在活跃推送，MetaGPT 仓侧放缓——国产线内部也分活跃/放缓。Same-category contrast: both are China-origin multi-agent frameworks — AgentScope (Alibaba line) is still pushing actively while MetaGPT's repo has slowed; even within the China-origin line there is an active/slow split.
- DevinDevin跨大类对照：Devin 是替你干活的成品（coding 大类），MetaGPT 是让你搭「软件公司流水线」的底座（framework 大类）——§7 收录标准的正反两面。Cross-category contrast: Devin is a finished product that works for you (coding category); MetaGPT is a base for you to build a "software-company pipeline" (framework category) — the two sides of the §7 inclusion standard.
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.