Jan
Jan is an open-source desktop app under Apache-2.0 that its README calls an open-source ChatGPT replacement. It downloads open models such as Llama, Gemma, Qwen and gpt-oss from Hugging Face and runs them on your own computer, connects to OpenAI, Anthropic, Mistral and Groq when you choose, serves a local OpenAI-compatible API for other apps, and supports MCP. It runs on Windows, macOS and Linux.
What does Jan do?
Jan's README is titled "Open-source ChatGPT replacement", and its tagline is that it is "bringing the best of open-source AI in an easy-to-use product." It downloads and runs models such as Llama, Gemma, Qwen and gpt-oss from Hugging Face on your own machine, and it can also "connect to GPT models via OpenAI, Claude models via Anthropic, Mistral, Groq, MiniMax, and others." Its privacy line is "Everything runs locally when you want it to."
It serves an OpenAI-compatible API at localhost:1337, so other applications can use the models Jan is running. MCP integration adds agent tools, and you can build custom assistants for specific tasks. It is built on llama.cpp and Tauri, with builds for CPU, Vulkan, Metal, CUDA 12 and 13, and ROCm. The README does not describe document chat, web search, image generation, voice or multi-user features.
Key facts
- Licence: Apache-2.0. The LICENSE file is a short Menlo Research notice followed by the Apache notice, and it adds one line: "Attribution is requested in user-facing documentation and materials, where appropriate", a request rather than a condition. GitHub reports it as Other.
- Local models from Hugging Face, such as Llama, Gemma, Qwen and gpt-oss; cloud models through OpenAI, Anthropic, Mistral, Groq, MiniMax and others.
- An OpenAI-compatible API server at localhost:1337 for other applications.
- MCP integration and custom assistants.
- Built on llama.cpp and Tauri, with CPU, Vulkan, Metal, CUDA 12, CUDA 13 and ROCm builds.
- Minimum specs from the README for macOS 13.6 or later: 8 GB of RAM for 3B models, 16 GB for 7B, 32 GB for 13B.
- Install: jan.exe, jan.dmg, jan.deb or jan.AppImage, the Microsoft Store, or Flathub.
- Latest tagged release when read: v0.8.4 on 2026-07-22; 44,516 stars on 2026-09-17.
What does it replace, and where does it fall short?
Jan replaces ChatGPT for one person on one computer: install it, pick an open model, and chat with nothing leaving the machine, with the option to switch to a hosted model when you need one. Its local server at port 1337 also stands in for the OpenAI API for other apps on the same machine, which is why it appears on the open-source alternatives to the OpenAI API page too.
Where it falls short: the README describes no document chat, web search, image generation or voice, which ChatGPT has, and no multi-user features. How good a local model is depends on your hardware: the README's own guide is 8 GB of RAM for a 3B model up to 32 GB for 13B. For a desktop client with document chat and more features, AnythingLLM.
How does Jan run?
Download the installer for Windows, macOS or Linux, or install from the Microsoft Store or Flathub. Pick a model to download, or add an API key for a hosted provider. The local API starts at localhost:1337. Building from source needs Node.js 20 or later, Yarn, Make and Rust. The README describes no hosted or paid tier; business enquiries go to hello@jan.ai.
Who is Jan for?
Someone who wants a private ChatGPT on their own laptop, with open models and no account, and a local API other tools can call. A team that needs sign-in, shared documents and roles should use Open WebUI or LibreChat.
What limits does the README state?
From the README: minimum specs on macOS 13.6 or later are 8 GB of RAM for 3B models, 16 GB for 7B and 32 GB for 13B; Windows CUDA builds can exceed the 260-character path limit; building on Windows does not work from a plain cmd.exe.
Questions people ask
Is Jan open source?
Yes. The licence is Apache-2.0. Its LICENSE file adds one line asking for attribution in user-facing materials where appropriate, which is a request rather than a condition, so the grant is plain Apache-2.0. GitHub shows Other only because the file is a short notice rather than the full licence text.
Does Jan work offline?
Yes. It downloads open models from Hugging Face and runs them on your computer through llama.cpp, and the README says everything runs locally when you want it to. Hosted providers such as OpenAI and Anthropic are optional, added with your own key.
What computer do I need to run Jan?
The README's minimum for a decent experience on macOS 13.6 or later is 8 GB of RAM for 3B models, 16 GB for 7B models and 32 GB for 13B models. Builds exist for CPU, Vulkan, Apple Metal, NVIDIA CUDA and AMD ROCm, on Windows 10 or later, macOS and Linux.
How does Jan compare with ChatGPT?
Jan gives you a ChatGPT-style chat on your own computer with open models and no account, plus a local API. ChatGPT offers OpenAI's models with search, images and voice built in. The open-source alternatives to ChatGPT page compares Jan with Open WebUI, LibreChat, AnythingLLM and the other clients.
Can other apps use the models running in Jan?
Yes. Jan runs an OpenAI-compatible API server at localhost:1337, so any tool that talks to the OpenAI API can point at it instead. That is the same role Ollama and LocalAI play, with Jan's chat window on top.
Sources
- Jan README and LICENSE: github.com/janhq/jan, read 2026-09-17.
Compared with the others
On the open-source alternatives to ChatGPT page, Jan is the pick for a private chatgpt on one laptop. Apache-2.0 desktop app that runs open models from Hugging Face on your computer, connects to hosted models when you choose, and serves a local OpenAI-compatible API.
Also on that page: Open WebUI for a private chatgpt for a whole team, LibreChat for every provider in one mit platform, AnythingLLM for chatting with your documents.
On the open-source alternatives to the OpenAI API page, Jan is the pick for a local api with a chat window on top. Apache-2.0 desktop app that runs open models and serves an OpenAI-compatible API at localhost:1337 for other apps.
Also on that page: Ollama for the quickest local start, LocalAI for every modality behind openai-compatible apis, llama.cpp for the engine on any hardware.
More chat interfaces
A self-hosted chat platform for Ollama and OpenAI-compatible models with document search over 9 vector databases, web search, image generation, roles and single sign-on, under a BSD-3 licence with an added branding clause.
LobeHub, formerly LobeChat, is an agent workspace with an agent builder, agent groups, scheduled runs, editable memory and a library of over 10,000 tools and MCP plugins, under a community licence that needs a paid licence to distribute a modified version.
An MIT all-in-one AI app from Mintplex Labs for chatting with your documents with citations, running agents with MCP, and using any of 39 model providers including local llama.cpp and Ollama, as a desktop app or a multi-user Docker server.
An AGPL-3.0 desktop client for Windows, macOS and Linux that connects OpenAI, Gemini, Anthropic and local Ollama or LM Studio models, with 300+ preset assistants, side-by-side model chats, file support and MCP servers.
An AGPL-3.0 desktop app for local models, formerly text-generation-webui, with five engines including llama.cpp and ExLlamaV3, an OpenAI and Anthropic-compatible API, tool calling, file chat, image generation and LoRA training, fully offline.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/janhq/jan.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.