Jan
Jan’s Goal
Section titled “Jan’s Goal”Jan’s goal is to build superintelligence that you can self-host and use locally.
What is Jan?
Section titled “What is Jan?”Jan is an open-source AI ecosystem that runs on your hardware. We’re building towards open superintelligence - a complete AI platform you actually own.
The Ecosystem
Section titled “The Ecosystem”Models: We build specialized models for real tasks, not general-purpose assistants:
- Jan-Nano (32k/128k): 4B parameters designed for deep research with MCP. The 128k version processes entire papers, codebases, or legal documents in one go
- Lucy: 1.7B model that runs agentic web search on your phone. Small enough for CPU, smart enough for complex searches
- Jan-v1: 4B model for agentic reasoning and tool use, achieving 91.1% on SimpleQA
We also integrate the best open-source models - from OpenAI’s gpt-oss to community GGUF models on Hugging Face. The goal: make powerful AI accessible to everyone, not just those with server farms.
Applications: Jan Desktop runs on your computer today. Web, mobile, and server versions coming in late 2025. Everything syncs, everything works together.
Tools: Connect to the real world through Model Context Protocol (MCP). Design with Canva, analyze data in Jupyter notebooks, control browsers, execute code in E2B sandboxes. Your AI can actually do things, not just talk about them.
Core Features
Section titled “Core Features”Run Models Locally
Section titled “Run Models Locally”- Download any GGUF model from Hugging Face
- Use OpenAI’s gpt-oss models (120b and 20b)
- Automatic GPU acceleration (NVIDIA/AMD/Intel/Apple Silicon)
- OpenAI-compatible API at
localhost:1337
Connect to Cloud (Optional)
Section titled “Connect to Cloud (Optional)”- Your API keys for OpenAI, Anthropic, etc.
- Jan.ai cloud models (coming late 2025)
- Self-hosted Jan Server (soon)
Extend with MCP Tools
Section titled “Extend with MCP Tools”Growing ecosystem of real-world integrations:
- Creative Work: Generate designs with Canva
- Data Analysis: Execute Python in Jupyter notebooks
- Web Automation: Control browsers with Browserbase and Browser Use
- Code Execution: Run code safely in E2B sandboxes
- Search & Research: Access current information via Exa, Perplexity, and Octagon
- More coming: The MCP ecosystem is expanding rapidly
Architecture
Section titled “Architecture”Jan is built on:
- Llama.cpp for inference
- Model Context Protocol for tool integration
- Local-first data storage in
~/jan
Why Jan?
Section titled “Why Jan?”| Feature | Other AI Platforms | Jan |
|---|---|---|
| Deployment | Their servers only | Your device, your servers, or our cloud |
| Models | One-size-fits-all | Specialized models for specific tasks |
| Data | Stored on their servers | Stays on your hardware |
| Cost | Monthly subscription | Free locally, pay for cloud |
| Extensibility | Limited APIs | Full ecosystem with MCP tools |
| Ownership | You rent access | You own everything |
Development Philosophy
Section titled “Development Philosophy”- Local First: Everything works offline. Cloud is optional.
- User Owned: Your data, your models, your compute.
- Built in Public: Watch our models train. See our code. Track our progress.
System Requirements
Section titled “System Requirements”Minimum: 8GB RAM, 10GB storage
Recommended: 16GB RAM, GPU (NVIDIA/AMD/Intel), 50GB storage
Supported: Windows 10+, macOS 12+, Linux (Ubuntu 20.04+)
What’s Next?
Section titled “What’s Next?”When will mobile/web versions launch?
- Jan Web: Beta late 2025
- Jan Mobile: Late 2025
- Jan Server: Late 2025
All versions will sync seamlessly.
What models are available?
Jan Models:
- Jan-Nano (32k/128k): Deep research with MCP integration
- Lucy: Mobile-optimized agentic search (1.7B)
- Jan-v1: Agentic reasoning and tool use (4B)
Open Source:
- OpenAI’s gpt-oss models (120b and 20b)
- Any GGUF model from Hugging Face
Cloud (with your API keys):
- OpenAI, Anthropic, Mistral, Groq, and more
Coming late 2025:
- More specialized models for specific tasks
What are MCP tools?
MCP (Model Context Protocol) lets AI interact with real applications. Instead of just generating text, your AI can:
- Create designs in Canva
- Analyze data in Jupyter notebooks
- Browse and interact with websites
- Execute code in sandboxes
- Search the web for current information
All through natural language conversation.
How does Jan make money?
- Local use: Always free
- Cloud features: Optional paid services (coming late 2025)
- Enterprise: Self-hosted deployment and support
We don’t sell your data. We sell software and services.
Can I contribute?
Yes. Everything is open:
- GitHub - Code contributions
- Model Training - See how we train
- Discord - Join discussions
- Model Testing - Help evaluate models
Is this just another AI wrapper?
No. We’re building:
- Our own models trained for specific tasks
- Complete local AI infrastructure
- Tools that extend model capabilities via MCP
- An ecosystem that works offline
Other platforms are models behind APIs you rent. Jan is a complete AI platform you own.
What about privacy?
Local mode: Your data never leaves your device. Period.
Cloud mode: You choose when to use cloud features. Clear separation.
See our Privacy Policy.
Get Started
Section titled “Get Started”- Install Jan Desktop - Your AI workstation
- Download Models - Choose from gpt-oss, community models, or cloud
- Explore MCP Tools - Connect to real applications
- Build with our API - OpenAI-compatible at localhost:1337