different-ai/openwork
The open-source alternative to Claude Cowork (powered by opencode)
TypeScript · ★ 22,935 · Fork 2,267 · 在 GitHub 查看
在手机上浏览源码、订阅每日 AI 趋势日报与项目架构深度解读:获取 CopoHub 客户端(iOS & HarmonyOS)→
GitHub Trending 上榜记录
different-ai/openwork 已在 GitHub Trending 每日榜单上榜 6 天,最高排名第 2,首次上榜 2026-07-29,最近一次上榜 2026-08-03。
| 日期 | 排名 | Star | 当日增长 |
|---|---|---|---|
| 2026-08-03 | #7 | 20,614 | +280 |
| 2026-08-02 | #7 | 20,271 | +585 |
| 2026-08-01 | #2 | 19,696 | +806 |
| 2026-07-31 | #2 | 19,446 | +915 |
| 2026-07-30 | #4 | 18,656 | +916 |
| 2026-07-29 | #9 | 17,832 | +58 |
同期上榜的项目
- lyogavin/airllm – AirLLM 2.8T LLM inference with single 4GB GPU
- zhaoxuya520/reverse-skill – Reverse Engineering / Authorized Penetration Testing / Security Research Skill Router Pack AI-powered routing + On-demand toolchain bootstrapping + Self-evolving knowledge base Supports Claude Code, Kiro, Cursor, Cline, and other AI coding clients 逆向/渗透/安全技能路由包 - AI 自动路由 + 按需自举工具链 + 自动进化经验库 | 支持 Claude Code / Kiro / Cursor / Cline 等代码 AI 客户端
- firecrawl/pdf-inspector – Fast Rust library for PDF inspection, classification, and text extraction. Intelligently detects scanned vs text-based PDFs to enable smart routing decisions.
- esengine/DeepSeek-Reasonix – DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.
- TencentCloud/TencentDB-Agent-Memory – TencentDB Agent Memory is a team-level memory hub for AI Agents — turning conversations, docs, and code into four reusable memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) that are governed, shared, and equipped across agents and frameworks.
- microsoft/AI-For-Beginners – 12 Weeks, 24 Lessons, AI for All!
openwork 项目深度剖析与商业评估报告
openwork 项目深度剖析与商业评估报告
1. 核心定位与商业落地价值
根本痛点:AI Agent(如 Claude Code、Codex、Cursor)的碎片化问题。当前开发者面临“每个 Agent 有自己的工具、MCP、技能”的困境,导致跨工具、跨团队、跨机器的能力无法复用,形成严重的“AI 工具孤岛”。
商业落地场景: 企业 AI 治理平台:OpenWork Den 提供了组织级的 MCP 管理、模型提供商控制、技能发布和市场管理,直接对标企业级 AI 安全合规需求。企业可通过单一控制平面管理所有 Agent 的能力边界,避免影子 IT 风险。 AI 工作流共享市场:通过技能/插件的市场机制,实现“一次构建,到处运行”,降低企业 AI 资产沉淀和复用成本。 跨工具 Agent 编排:解决大型企业同时使用 Cursor、Claude Code、Codex 等工具的兼容性问题,统一能力出口。
竞品差异化与护城河: 与 Claude Cowork/Codex 相比,openwork 强调去中心化:Agent 本身不是必须的,用户可以通过 MCP 协议从任意兼容客户端接入。 与 LangChain/LlamaIndex 相比,openwork 更聚焦于桌面端到企业端的完整链路,提供从个人开发到组织治理的闭环,而非纯框架。 护城河在于其 MCP 生态的“网络效应”:用户越多,发布的技能/插件越多,对其他 Agent 的吸引力越强,形成正向循环。Den 控制平面进一步锁定企业客户。
2. 核心架构设计与技术深度
模块拆解思想: 三层架构:桌面客户端(Electron + TypeScript)→ MCP 协议层(HTTP Remote MCP Server)→ 控制平面(OpenWork Den)。 能力抽象:将“技能(Skills)”、“插件(Plugins)”、“MCP 连接”抽象为统一的“能力(Capability)”,通过 searchcapabilities 和 executecapability 两个工具暴露给 Agent,实现 LSP 风格的动态发现机制。 多 Profile 隔离:支持通过 OPENWORKDEVPROFILE 实现多工作区并行,每个 Profile 拥有独立的凭据、配置和状态,类似于 IDE 的工作区管理。
关键设计模式: MCP 作为适配器模式:将各类 Agent 的差异封装在 MCP 协议层,Agent 只需实现 MCP 客户端即可接入,无需修改 openwork 核心。 OAuth 委托认证:MCP 客户端通过 OAuth 流程获取令牌,Den 负责权限验证,实现“一次登录,全局授权”。 事件驱动的技能执行:executecapability 可能触发异步流水线,通过回调或轮询获取结果,支持长时间运行的任务(如代码审查、批量文档处理)。
在 README 中展示上榜徽章
[](https://github.fq6825.top/repo/different-ai/openwork)