// the find
janhq/jan
Jan is an open source alternative to ChatGPT that runs 100% offline on your computer.
Jan is a cross-platform desktop app for running LLMs locally, built with Tauri and TypeScript on top of llama.cpp. It exposes an OpenAI-compatible API at localhost:1337, so any tool that already speaks OpenAI (Cursor, Continue, LangChain) can use it as a drop-in offline backend. Aimed at developers and privacy-conscious users who want local inference without managing a server.
The localhost:1337 OpenAI-compatible server is the actual killer feature — you get local inference without rewriting any of your existing tooling. MCP integration is shipped and functional, which puts it ahead of most local-inference GUIs for agentic workflows. The engine abstraction in core/ (LocalOAIEngine, RemoteOAIEngine, OAIEngine) is clean enough that adding a new backend doesn't touch UI code. Ships to Microsoft Store and Flathub, which means non-technical users can install it without touching a terminal.
ARM64 Linux has no first-class binary — the README links to a workaround GitHub issue, not a build. The monorepo bundles a Next.js docs site, a Python autoqa harness, and a separate core/ npm package with its own build pipeline; onboarding a new contributor means understanding four different build systems. Model selection still requires knowing GGUF quantization tradeoffs (Q4_K_M vs Q8_0) — the UI downloads models but leaves users to figure out which variant won't OOM their machine. GPU acceleration details (CUDA/ROCm/Metal build variants) are not documented in the README, so Windows users with AMD cards are on their own.