Bottom line: Jan is a desktop AI chatbot that is completely free — not a free trial, not a credit drip. It is open source, runs AI models directly on your own computer, and works offline once a model is downloaded. The catch is the usual local-AI catch: your hardware is the limit, and the strongest cloud models are not included.
Details checked October 15, 2026, against Jan’s project documentation and repository. This is a research-based review; we do not claim hands-on testing.
What does Jan do?
Jan gives you a ChatGPT-style chat window that lives on your machine instead of a company’s server. You download an open model — options include families such as Llama, Gemma, Qwen, and Mistral, sourced through Hugging Face — and chat with it locally on Windows, macOS, or Linux. Conversations and files stay on your computer, and after the initial model download, no internet connection is needed.
Jan also covers the power-user bases: custom assistants, document chat, GPU acceleration, a local OpenAI-compatible API server (on localhost:1337) for other tools, and Model Context Protocol (MCP) support.
What is actually free with Jan?
All of it. Jan is free and open source under the AGPLv3 license, and the project states there is no paid version of the app. Downloading the app, downloading open models, and chatting locally cost nothing and have no usage caps — the meter is your own hardware.
The one place money can appear is optional. If you connect a cloud provider inside Jan — OpenAI, Anthropic, Mistral, Groq, and others are supported — you supply your own API key and that provider bills you for usage. Local models never touch that meter.
What are the limits of running AI locally?
- Hardware decides quality: small models run on modest machines (plan on at least 8 GB of RAM and around 10 GB of storage per model library); larger, smarter models want more RAM and a capable GPU.
- Speed varies: responses on an older laptop will be slower than any cloud chatbot.
- Model choice takes learning: picking sizes and quantizations is simpler than it used to be, but it is still one more decision than opening a web chatbot.
- No built-in live web: a local model knows only its training data unless you connect tools or cloud models yourself.
Pros and cons of Jan
Pros
- Completely free and open source, with no account and no paid version
- Private by design — chats can run 100% offline and never leave your machine
- One app for many models, plus optional cloud models through your own keys
- Local API and MCP support make it useful to developers, not just chatters
Cons
- Performance is capped by your computer; weak hardware means small models and slow replies
- Model downloads are large and eat disk space
- Open models trail the best cloud models on hard reasoning
Who is Jan best for?
Privacy-conscious users, writers who want AI help off-grid, and developers who want a local model endpoint without API bills. On an old machine with hard questions, a cloud chatbot’s free tier serves you better.
Frequently asked questions
Is Jan really free?
Yes. The app and local models are free and open source (AGPLv3). The project has no paid version. Only optional cloud models, used through your own API keys, cost money — paid to that provider, not to Jan.
Does Jan work offline?
Yes. Once a model is downloaded, local chat works with no internet connection at all.
What computer do I need for Jan?
A modern 64-bit machine with at least 8 GB of RAM is the practical floor; 16 GB or more and a dedicated GPU make larger models comfortable.
How is Jan different from LM Studio or Ollama?
All three run local models for free. Jan leans into a beginner-friendly chat app with custom assistants and MCP support; LM Studio leans into model testing, and Ollama into a developer-first command line and API.
Verdict: If “free” should mean free forever — no credits, no watermark, no account — Jan is the real article. Just bring hardware that can carry the model you want.