Jan
Open-source desktop app that runs LLMs 100% offline with a ChatGPT-familiar interface.
About
Jan is the polished answer to local AI for people who don't want a terminal. It is a cross-platform desktop app (Windows, macOS, Linux) with a clean ChatGPT-like interface that downloads and runs open models from Hugging Face with one click: Llama, Gemma, Qwen, Mistral, DeepSeek, GPT-OSS, and any GGUF model, with hardware-fit tagging so you know what your machine can handle.
Beyond chat, Jan ships a model Hub, web search and page fetching, image input, MCP integrations for agentic capabilities, custom assistants, and an OpenAI-compatible local API server on port 1337 that other tools can point at. It can also connect to cloud providers (OpenAI, Anthropic, Groq, Mistral) from the same interface when you want them.
Apache-2.0 licensed with around 45,000 GitHub stars and active development (v0.8.6 reviewed hands-on in October 2026). Independent reviews consistently land on the same verdict: the closest thing to a finished-product ChatGPT experience that never sends your data anywhere. Performance, as always with local inference, depends on your hardware.
What it can do
- One-click local model downloads from Hugging Face
- 100% offline chat with a ChatGPT-familiar UI
- OpenAI-compatible local API server (localhost:1337)
- MCP integrations, custom assistants, web search
- Optional cloud provider connections from the same UI
Setup
Download jan.ai for your OS (exe/dmg/deb/AppImage) or build from source with make dev; pick a model from the Hub and start chatting.
Prerequisites
Windows 10+, macOS 13.6+, or Linux; 8GB+ RAM for small models, 16-32GB for larger ones; GPU helps a lot.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 44,956 GitHub stars (janhq/jan); repo pushed 2026-10-15 (same-day activity); Apache-2.0 confirmed via LICENSE file; Offline-first chat, Hub model downloads, MCP support, and localhost:1337 OpenAI-compatible server verified against README and independent hands-on review (v0.8.6, Oct 2026). Manual review: local inference performance is hardware-bound -- the app is only as good as the machine it runs on; cloud connections are opt-in. License: Apache-2.0.
★ 44,956 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 44,956 GitHub stars (janhq/jan); repo pushed 2026-10-15 (same-day activity); Apache-2.0 confirmed via LICENSE file
- · Offline-first chat, Hub model downloads, MCP support, and localhost:1337 OpenAI-compatible server verified against README and independent hands-on review (v0.8.6, Oct 2026)
- · manual review: local inference performance is hardware-bound -- the app is only as good as the machine it runs on; cloud connections are opt-in
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.