zosma-cowork

作者 zosmaai

Desktop GUI for the pi coding agent — open-source Claude Cowork alternative

110
Stars
21
Forks
TypeScript
语言
2026/8/23
添加时间

⚠️ 第三方软件声明

本 Skill 为第三方开源软件,独立托管于 GitHub。SkillTip 仅为信息目录,不控制或维护底层仓库。所显示的安全检查为自动化且范围有限,安装前请自行审查源码。

阅读服务条款

安装

添加到你的 Claude Code skills 目录:

# Add to your Claude Code skills
git clone https://github.com/zosmaai/zosma-cowork

快速入门

使用 zosma-cowork 等 Skills 的指南。

安全报告

已验证

上次扫描:—

{
  "status": "PASSED",
  "issues": []
}

README.md

Zosma Cowork 🇮🇳

Zosma Cowork

English | 中文 | Español | 日本語 | Deutsch | Français | Português | Русский | 한국어 | हिंदी

CI Release License: MIT Platforms Built with Tauri Discord GitHub Repo Stars

📖 Documentation: cowork.zosma.ai


A desktop agentic work harness built on pi, the minimal, language-agnostic coding agent harness. Streaming, thinking, tool calls, multi-turn sessions — all free, all open-source, all local.

Built by Zosma AI.

Gallery

Zosma Cowork demo Zosma Cowork screenshot

Invoice processing with natural language agents. See more demos at zosma.ai/zosma-cowork/gallery

Why Zosma Cowork?

🌟 Built on pi

Zosma Cowork is a desktop application built on pi — the minimal, language-agnostic coding agent harness. pi's philosophy of simplicity and composability carries directly into your desktop experience. Every pi extension works out of the box, with zero wrappers or adapters.

🆓 Free & Open Source

Zosma Cowork is 100% free and open-source (MIT). Bring your own API key, use an existing subscription (Claude, ChatGPT, Copilot), or run local models — you stay in control.

🧩 Full pi Extension Ecosystem

The pi ecosystem includes hundreds of extensions, skills, tools, prompts, and themes — all compatible with Zosma Cowork. Plug them into your ~/.zosmaai/cowork/ directory and they just work. No wrapping, no porting, no lock-in.

Features

  • Node.js agent sidecar — The pi-mono TypeScript SDK runs in a managed sidecar process for full agent capabilities (extensions, tools, providers)
  • Thin Tauri relay — The Rust layer is a minimal IPC bridge between React and the sidecar, keeping the native desktop shell lightweight
  • pi extension ecosystem — Compatible with pi extensions via DefaultResourceLoader — skills, tools, and prompts auto-discovered from ~/.zosmaai/cowork/
  • Multi-turn sessions — Full conversation continuity with persistent session history
  • Streaming responses — See the agent think, write, and call tools in real-time
  • Thinking blocks — Expandable reasoning from the model
  • Tool call timeline — Live bash/edit/write tool calls with args and results
  • Session management — Persistent chat sessions saved to ~/.zosmaai/cowork/
  • Light & dark mode — Warm cream light mode, warm charcoal dark mode
  • Keyboard shortcutsCmd/Ctrl+Shift+K to focus, Cmd/Ctrl+N for new session
  • Abort & steering — Stop a running agent mid-turn, send follow-up steering messages
  • Claude-inspired UI — 3-column layout with sidebar, workspace, and info panel

Architecture

Zosma Cowork architecture diagram
Edit this diagram

The diagram is generated from assets/architecture.mmd. To update:

# Edit assets/architecture.mmd, then re-render:
mmdc -i assets/architecture.mmd -o assets/architecture.png -t default -b white -w 900 -H 700 -s 2

See the full Product Roadmap for phased plans, competitive positioning, and technical decisions.

Architecture diagrams (6 views)

System Architecture — Full stack: React → Rust → Node sidecar → pi SDK

System Architecture

Communication Flow — Sequence diagram: streaming events pipeline

Communication Flow

Extension Ecosystem — 4 tiers: installed, vendored, planned, future

Extension Ecosystem

Competitive Positioning — Quadrant chart vs Cursor, Hermes, Slack AI

Competitive Positioning

Roadmap Timeline — Gantt chart: 6 phases, 28 weeks

Roadmap Timeline

Data Flow — User → App → Agent → Storage layers

Data Flow

Tech Stack

LayerTechnology
FrontendReact 19, Tailwind CSS v4, Radix UI
Desktop ShellTauri v2, Rust, Tokio
Agent EngineNode.js sidecar using @earendil-works/pi-coding-agent (pi-mono SDK)
TestingVitest, Testing Library, jsdom
LintingBiome (frontend + sidecar), Clippy (Tauri relay)

Development

Prerequisites

Quick Start

# Install frontend dependencies
npm install

# Install agent-sidecar dependencies
cd agent-sidecar && npm install && cd ..

# Run frontend dev server
npm run dev:frontend

# Run full Tauri app (frontend + Rust relay + Node.js sidecar)
npm run dev

npm run dev runs the sidecar from TypeScript source via tsx — no bundle needed. On a fresh checkout it auto-generates lightweight dev stubs for the Tauri bundle resources (src-tauri/agent-sidecar/index.cjs, src-tauri/binaries/node) so the Rust shell can compile. The real sidecar bundle and Node.js binary are produced only by the production build (npm run build).

Scripts

# Frontend
npm run lint          # Biome lint
npm run typecheck     # TypeScript check
npm run test          # Vitest run
npm run validate      # lint + typecheck + test
npm run format        # Biome format

# Tauri
npm run build:frontend
npm run build         # Build release binary

# Agent Sidecar
cd agent-sidecar
npm run build         # TypeScript → JavaScript
npm run dev           # tsx watch (standalone development)

# Rust (Tauri relay only)
cargo fmt --all --check
cargo clippy --workspace -- -D warnings

Staging builds

Every merge to main produces unsigned cross-platform installers via the Staging Build workflow (.github/workflows/staging-build.yml). The bundles are attached as workflow artifacts (14-day retention) and a Discord embed with auth-free nightly.link download URLs is posted to whatever channel the DISCORD_STAGING_WEBHOOK repo secret points at (typically #staging-builds).

This flow does not create a GitHub Release, tag a commit, or publish to AUR / winget / Homebrew — those side-effects remain gated on the tag-triggered release.yml. See issue #133 for the design.

To run a staging build on demand, trigger Staging Build from the Actions tab via Run workflow.

One-time setup: in Discord, open the target channel → Edit Channel → Integrations → Webhooks → New Webhook, copy the URL, then add it as a GitHub repo secret named DISCORD_STAGING_WEBHOOK (Settings → Secrets and variables → Actions → New repository secret). The workflow degrades gracefully — if the secret is unset the notify job emits a warning and exits 0, so the build itself still succeeds and the artifacts are still uploaded.

Config & Data

WhatLocationNotes
LLM providers & API keys~/.zosmaai/cowork/auth.jsonManaged by the app
Model definitions~/.zosmaai/cowork/models.jsonManaged by the app
Extensions & skills~/.zosmaai/cowork/extensions/Pi-compatible extensions
Session history~/.zosmaai/cowork/Managed by Zosma Cowork

IPC Protocol

The Tauri relay communicates with the Node.js sidecar via stdin/stdout JSON lines:

Commands (→ sidecar):

CommandDescription
initInitialize agent with zosmaDir config
get_modelsList available models from all providers
promptSend user message, stream events
abortCancel running prompt
set_modelSwitch active model
save_authSave API key for a provider
reloadReinitialize with fresh extensions/auth

Events (← sidecar):

EventUI Effect
readyModels loaded, enable UI
eventAgent session events (thinking, text, tool calls)
donePrompt completed
resultResponse to a request command
errorError with message

Project Structure

zosma-cowork/
├── agent-sidecar/                # Node.js agent process
│   └── src/
│       └── index.ts              # Sidecar: pi-mono SDK, stdin/stdout protocol
├── src/                          # React frontend
│   ├── components/               # UI components
│   │   ├── ChatMessage.tsx       # Message with thinking + tool calls
│   │   ├── ThinkingBlock.tsx     # Expandable reasoning
│   │   ├── ToolCallTimeline.tsx  # Tool execution timeline
│   │   ├── MessageInput.tsx      # Chat input
│   │   └── ui/                   # Primitives (tooltip, badge, etc.)
│   ├── hooks/
│   │   ├── usePiStream.ts        # Streaming state machine (useReducer)
│   │   └── useSessions.ts        # Session persistence
│   ├── types/
│   │   ├── index.ts              # ChatMessage, ToolCallInfo
│   │   └── pi-events.ts          # CoworkEvent types
│   ├── App.tsx                   # Main 3-column layout
│   └── App.css                   # Tailwind theme (light + dark)
├── src-tauri/                    # Tauri desktop shell (thin Rust relay)
│   └── src/
│       ├── main.rs               # Entry point
│       └── lib.rs                # IPC commands → sidecar process
├── docs/                         # Architecture, plans & DESIGN.md (design system)
└── .github/workflows/            # CI/CD

Building UI? Follow the Design Guidelines — the global standard for Zosma's elevated blue-glass design system.

Thank you for starring Zosma Cowork!
If you find Zosma Cowork useful, ⭐ star the repo — it lets us know we're building something that matters.

Contributors

Contributors

🇮🇳 Made in India

Zosma Cowork — built from India by ZOSMAAI SOLUTIONS PRIVATE LIMITED.

Citation

If you use Zosma Cowork in your work, please cite it:

@software{zosma_cowork,
  author  = {Nayak, Arjun and Mhaskar, Akshay and Shanvit and Mishra, Devendra},
  title   = {{Zosma Cowork: A Desktop Agentic Work Harness}},
  url     = {https://github.com/zosmaai/zosma-cowork},
  version = {0.15.2},
  year    = {2026}
}

License

MIT © Zosma AI

常见问题

What is zosma-cowork?

zosma-cowork is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by zosmaai. Desktop GUI for the pi coding agent — open-source Claude Cowork alternative. It has 110 GitHub stars.

Is zosma-cowork safe to use?

zosma-cowork returned warnings in SkillsLLM's automated security scan. It has no critical vulnerabilities, but review the flagged issues in the Security Report section before adding it to your workflow.

How do I install zosma-cowork?

Clone the repository with "git clone https://github.com/zosmaai/zosma-cowork" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is zosma-cowork written in?

zosma-cowork is primarily written in TypeScript. It is open-source under zosmaai on GitHub, so you can review or fork the full source.

Are there alternatives to zosma-cowork?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh zosma-cowork against similar tools.

评论 (0)

暂无评论,成为第一个分享想法的人!

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情

claude-code

by anthropics

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

120,03119,897Shell
AI 智能体
查看详情

开发者还喜欢

基于喜欢此 Skill 的开发者投票和收藏

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

n8n

by n8n-io

12

Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.

201,88160,308TypeScript
MCP 服务器apisai-tools
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情