Kuse Cowork：An Open-Source, Model-Agnostic Alternative to Claude Cowork

March 26, 2026 · View on GitHub

Kuse Cowork：An Open-Source, Model-Agnostic Alternative to Claude Cowork

Kuse Cowork is a lightweight, open-source desktop cowork agent built for people who want local-first execution, full model freedom, and real privacy control.

Works with any models, BYOK, written in Rust 🚀

Demo video: Kuse Cowork in action

It is an open-source desktop cowork agent created by Kuse, an AI document generator & Presentation Maker from your knowledge base. Transform docs, PDFs, YouTube, and images into formatted docs, infographics, mind maps, and flashcards—instantly. Stunning, professional, and ready to use.

✨ Why Kuse Cowork?

🔐 BYOK (Bring Your Own Key)

Use your own API keys or even bring your own local models for ultimate privacy control.

⚡ Pure Rust Agent

Agent fully written in Rust with zero external dependencies - blazingly fast and memory-safe.

🌍 Native Cross-Platform

True native performance on macOS, Windows, and Linux.

🛡️ Container Isolation & Security

Uses Docker containers for secure command execution and complete isolation.

🧩 Extensible Skills System

Support for custom skills to extend agent capabilities. Default skills are: docx, pdf, pptx, xlsx.

🔗 MCP Protocol Support

Full support for Model Context Protocol (MCP) for seamless tool integration.

📰 News & Updates

Standard release

[2026-01-26] Release of v0.0.2: Fixing issue where "start task" button actively needs the user to set a cloud-based api key.

Experimental

[2026-01-29] Kuse_cowork now supports basic operations on Docx with an integrated UI && support working trace tracking.
[2026-01-26] Kuse_cowork now supports basic operations on Excel with an integrated UI.

🚀 Features

🔒 Local & Private: Runs entirely on your machine, API calls go directly to your chosen provider
🔑 BYOK Support: Use your own Anthropic, OpenAI, or local model APIs
🎯 Model Agnostic: Works with Claude, GPT, local models, and more
🖥️ Cross-Platform: macOS (ARM & Intel), Windows, and Linux
🪶 Lightweight: ~10MB app size using Tauri
🐳 Containerized: Docker isolation for enhanced security
🧩 Skills: Extensible skill system for custom capabilities
🔗 MCP: Model Context Protocol support for tool integration

Security Note

This is still an early project and please be super careful when connecting with your local folders.

🚀 Quick Start

Get up and running in minutes:

1. Build the project and start

Will update to a clean release build soon.

2. ⚙️ Configure Your AI Model

Open Settings (gear icon in sidebar)
Choose your AI provider:
- Anthropic Claude - Enter your Claude API key
- OpenAI GPT - Enter your OpenAI API key
- Local Models - Configure Ollama/LM Studio endpoint
Select your preferred model (Claude 3.5 Sonnet, GPT-4, etc.)

3. 🔑 Enter API Key

Add your API key in the settings
Keys are stored locally and never shared

4. 📁 Set Workspace Folder

Click "Select Project Path" when creating a new task
Choose your project folder or workspace directory
The agent will work within this folder context

5. 🎯 Start Your First Task!

Click "New Task"
Describe what you want to accomplish
Watch the AI agent work on your project
Review the plan and implementation steps

Example tasks:

"Organize my folders"
"Read all the receipts and make an expense reports"
"Summarize the meeting notes and give me all the TODOs."

🛠️ Development

Prerequisites

Node.js 18+
Rust (for Tauri)
Docker Desktop (required for container isolation)
Tauri Prerequisites

Note: Docker Desktop must be installed and running for container isolation features. Without Docker, the app will still work but commands will run without isolation.

Setup

# Clone the repo
git clone https://github.com/kuse-ai/kuse-cowork.git
cd kuse-cowork

# Install dependencies
npm install

# Run in development mode
npm run tauri dev

# Build for production
npm run tauri build

Project Structure

kuse-cowork/
├── src/                    # Frontend (SolidJS + TypeScript)
│   ├── components/         # UI components
│   ├── lib/               # Utilities (API clients, MCP)
│   └── stores/            # State management
├── src-tauri/             # Backend (Rust + Tauri)
│   ├── src/               # Rust source code
│   │   ├── agent/         # Agent implementation
│   │   ├── tools/         # Built-in tools
│   │   ├── skills/        # Skills system
│   │   ├── mcp/           # MCP protocol support
│   │   └── database.rs    # Local data storage
│   ├── Cargo.toml         # Rust dependencies
│   └── tauri.conf.json    # Tauri configuration
├── .github/workflows/     # CI/CD for cross-platform builds
└── docs/                  # Documentation and assets

🔧 Configuration

API Providers

Kuse Cowork supports multiple AI providers:

Anthropic Claude: Direct API integration
OpenAI GPT: Full GPT model support
Local Models: Ollama, LM Studio, or any OpenAI-compatible endpoint
Custom APIs: Configure any compatible endpoint

Settings

All settings are stored locally and never shared:

API Configuration: Keys and endpoints for your chosen provider
Model Selection: Choose from available models
Agent Behavior: Temperature, max tokens, system prompts
Security: Container isolation settings
Skills: Enable/disable custom skills
MCP Servers: Configure external tool providers

🛡️ Security & Privacy

Container Isolation

Kuse Cowork uses Docker containers to isolate all external command execution:

Complete isolation from your host system
Secure networking with controlled access
Resource limits to prevent abuse
Clean environments for each execution

Privacy First

No telemetry - nothing is sent to our servers
Local storage - all data stays on your machine
Direct API calls - communications only with your chosen AI provider
Open source - full transparency of all code

📄 License

MIT License - see LICENSE for details.

🏗️ Built With

Tauri - Lightweight desktop framework
Rust - Systems programming language

🚧 Roadmap & TODOs

Upcoming Features

📦 Streamlined Release Pipeline - Automated builds and easier distribution
🎯 Simplified Setup - One-click installation for non-developers
🐬 Lightweight Sandbox - Migrate to an lightweight sandbox.
🧠 Context Engineering - Enhanced support for better context management
🔧 Auto-configuration - Intelligent setup for common development environments
📱 Mobile Support - Cross-platform mobile app support