NoWork

Open-source desktop AI agent workspace for coding, research, documents, and multi-agent workflows.

NoWork lets you create AI workers like chat contacts. Each worker can plan, code, review, research, process files, and run workflows inside a local desktop app.

简体中文

---

📸 Screenshots

🖥️ Main Interface

A clean, contact-style workspace for managing multiple agent conversations.

🚀 Example: Building a Pomodoro Timer

A coding agent plans and builds a simple Pomodoro timer from scratch.

👥 Team Worker: Multi-Agent Collaboration

Team workers coordinate multiple agents — watch member activities, tool calls, and progress in real time.

📚 Knowledge Base: Wiki-Driven Knowledge Management

Build a structured Wiki from your documents — auto-ingest files, browse pages, search, and visualize knowledge graphs.

---

What is NoWork?

NoWork is a desktop-first multi-agent app built for practical work:

• Coding — planning, implementation, review, debugging
• Document work — Word, Excel, PowerPoint, PDF
• Research — web browsing, summarization, technical exploration
• Ongoing workflows — long conversations, scheduled runs, reusable workers

Instead of juggling prompts, terminals, and scripts, you talk to workers as if they were contacts in a chat app.

---

🤔 Why NoWork?

Command-line AI tools are powerful, but they still create friction for many users. Existing desktop wrappers often feel unstable, too thin, or too dependent on manual environment setup.

NoWork takes a different approach:

• Desktop-first, not terminal-first — install and run without needing CLI workflows
• Chat-native interaction — every agent or team is a contact, so delegation feels natural
• Bundled runtime — the app ships with its own backend and embedded Python runtime
• Local workspace access — agents can work with real files inside configured directories
• Small-team and solo-friendly — practical multi-agent automation without platform complexity

The goal is simple: let AI handle repetitive work, so you can stay focused on decisions and results.

---

✨ Key Capabilities

Capability	Description
Contact-style workers	Each Agent / Team appears as a chat contact. Click and start working.
Multi-agent collaboration	Built-in planning, coding, review, architecture, research, and document workers.
50+ model providers	Powered by Agno. Supports OpenAI, Anthropic, Google, DeepSeek, Qwen, Ollama, vLLM, and many more.
Long-session continuity	Automatic context compaction helps preserve useful state in long conversations.
Workspace-safe file access	Sandboxed file and shell operations limited to configured directories and permissions.
Document processing skills	Built-in support for Word, Excel, PowerPoint, and PDF workflows.
Knowledge bases	Wiki-driven knowledge management inspired by Karpathy's LLM Wiki pattern, with implementation reference from llm_wiki. Supports auto-ingest, full-text search, knowledge graphs, and Worker integration. Works in both Wiki mode and traditional vector mode.
Scheduled runs	Set recurring tasks for workers to run automatically.
MCP integration	Connect external tools through Model Context Protocol.
Desktop-native packaging	Tauri desktop shell with bundled backend and embedded Python runtime.
Bilingual UI	English / 简体中文 support with instant switching.

---

🚀 Quick Start

For end users

1. Download the latest release from the Releases page
2. Install the desktop app
3. Configure a model provider
4. Choose a worker
5. Start chatting and delegating work

Minimal model configuration

Create a provider file such as:

# server/config/models/my-provider.yaml
provider_id: my-provider
name: My Provider
base_url: https://api.openai.com/v1
api_key: sk-xxx
models:
  - id: gpt-4o
    name: GPT-4o
    image: true
    video: false

Then set the default model:

# server/config/config.yaml
default_model: my-provider/gpt-4o

After that, open the app, pick a worker, and start chatting.

---

🧠 Built-in Workers

NoWork ships with a practical default workforce for common engineering and document tasks.

Worker	Role	Type
Code Agent	Write, edit, and debug code	Agent
Planning Engineer	Analyze requirements and write implementation plans	Agent
Implementation Engineer	Execute plans step by step	Agent
Architecture Reviewer	Read-only review and risk analysis	Agent
Code Explorer	Search, navigate, and understand codebases	Agent
Documentation Researcher	Write docs and research technical topics	Agent
Doc Agent	Process Office/PDF documents and perform web research	Agent
Product R&D Team	Planner → coder → reviewer pipeline	Team

Workers are YAML-driven — add, remove, or customize them without changing application code.

---

⚙️ Configuration

File	Purpose
server/config/config.yaml	Global config: default model, tools, server settings
server/config/models/*.yaml	Model provider definitions, API keys, endpoints, capabilities
server/config/workers/*.yaml	Worker definitions: instructions, tools, workspaces, history, learning
server/config/knowledge/*.yaml	Knowledge base definitions: paths, wiki mode, purpose, auto sync
server/config/mcp.yaml	MCP server connections

Example worker config

# server/config/workers/my-worker.yaml
agent:
  id: my-worker
  name: My Worker
  instructions: You are a helpful coding assistant.
tools:
  - module: app.tools.codingTools
    class: CodingTools
    config:
      base_dirs:
        - C:/Users/me/projects
workspaces:
  - path: C:/Users/me/projects
    permission: read-write

---

🏗️ Architecture

NoWork uses a Tauri desktop shell, a React frontend, and a local FastAPI + Agno backend with an embedded Python runtime.

┌─ Tauri Desktop (Rust) ─────────────────────────────┐
│                                                    │
│  ┌─ React Frontend (Vite + TypeScript) ──────────┐ │
│  │  NavRail │ Worker List │ Chat Workspace       │ │
│  │  Voice   │ Schedules   │ Settings │ Help      │ │
│  └───────────────────────────────────────────────┘ │
│                      HTTP / SSE                    │
│  ┌─ Python Backend (FastAPI) ───────────────────┐ │
│  │  Agno AgentOS ─ Agent / Team / Workflow      │ │
│  │  CodingTools (sandboxed shell + file ops)    │ │
│  │  CompactionManager (context management)      │ │
│  │  MCP Client │ Knowledge Base │ Skills        │ │
│  │  Session Persistence (SQLite)                │ │
│  └──────────────────────────────────────────────┘ │
│  Tauri manages backend lifecycle automatically    │
└────────────────────────────────────────────────────┘

---

🛠️ Run in Development

Prerequisites

Tool	Version
Node.js	>= 18
Python	>= 3.10
Rust	stable

Setup

# Python deps
pip install -r requirements.txt

# Frontend deps
cd web
npm install
cd ..

Start development environment

powershell -ExecutionPolicy Bypass -File scripts/start-dev.ps1

Useful commands

# Frontend only
cd web
npm run dev

# Backend only
conda activate nowork
$env:PYTHONPATH = "server"
python -m app.run

# Full desktop app
npm run tauri:dev

---

📦 Build Release

# Recommended
powershell -ExecutionPolicy Bypass -File scripts/build-release.ps1

Output:

src-tauri/target/release/bundle/

Packaging notes

• Release builds bundle an embedded Python runtime
• Python dependencies are copied using a filtered allowlist, not a full Conda environment
• End users do not need Python, Conda, or manual dependency installation
• Server code and resources are packaged into the desktop app directly

---

📄 License

MIT License