Jan
Open-source ChatGPT alternative that runs 100% offline on your desktop
About
Jan is an open-source desktop AI assistant that runs LLMs entirely on your computer — no cloud, no API keys, no account. Built on Tauri, it ships for Windows, macOS, and Linux with a built-in model Hub of GGUF models (Llama, Gemma, Qwen, Mistral, DeepSeek, and more). When you do want cloud, it also connects to OpenAI, Anthropic, Mistral, Groq and others, and it exposes an OpenAI-compatible local API server on port 1337 so other tools can use your local models. It supports custom assistants, MCP integrations, and extensions. Apache-2.0 licensed, ~45k GitHub stars, actively developed (v0.8.6, Oct 2026).
What it can do
- **100% offline chat** — download GGUF models from the in-app Hub and run them locally via llama.cpp/MLX backends.
- **OpenAI-compatible local server** — `localhost:1337` with `/v1/chat/completions`, usable from Claude Code, Cursor, Windsurf, or any OpenAI SDK.
- **Cloud providers** — connect OpenAI, Anthropic, Mistral, Groq, OpenRouter via API keys; hybrid local/cloud in one UI.
- **MCP integration** — Model Context Protocol support for agentic tool use.
- **Custom assistants** — create specialized assistants for your tasks.
- **Extensions** — plugin system for customizing Jan.
Setup
Desktop (easiest)
- Download
jan.exe(Windows),jan.dmg(macOS universal), orjan.deb/jan.AppImage(Linux) from jan.ai or GitHub releases and install. - Open the Hub tab, pick a GGUF model (e.g. a small Qwen/Mistral build for modest machines), and download it inside the app.
- Start a New Chat and chat offline — no internet required after the model is downloaded.
Local OpenAI-compatible server
Jan exposes http://localhost:1337/v1/chat/completions — point any OpenAI SDK/client at it (API key: blank or local). Verified by independent tutorials.
From source
```bash
git clone https://github.com/janhq/jan && cd jan
make dev # Node >= 20, Yarn >= 4.5.3, Make, Rust (Tauri) required
```
Prerequisites
- A desktop OS: Windows (x64), macOS (universal), or Linux (deb/AppImage; ARM64 via community how-to). - Enough RAM/VRAM for the model you pick (3B models for modest machines, 7–13B for strong PCs). - No model is bundled — you download one from the in-app Hub first (needs internet once). - Build-from-source additionally needs Node ≥ 20, Yarn ≥ 4.5.3, Make, Rust; on Apple Silicon, the MetalToolchain Xcode component.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 44,913 GitHub stars (log-scaled vs catalog peers); last push 0d ago; 15 releases/12mo; 565 open issues. Manual review: repo moved menloresearch/jan → janhq/jan; 'zero telemetry' provider-claimed, not audited. License: Apache-2.0.
★ 44,913 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.6/10
- Maintenance · recency, releases, issue load
- 9.5/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 44,913 GitHub stars (log-scaled vs catalog peers)
- · last push 0d ago; 15 releases/12mo; 565 open issues
- · manual review: repo moved menloresearch/jan → janhq/jan; 'zero telemetry' provider-claimed, not audited
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.