What Jan does
Jan is an open-source, local-first desktop AI app for downloading and running models, connecting optional cloud providers, using MCP tools, and serving a local compatible API.
Jan packages local model discovery, chat, configuration, extensions, and an API into an open desktop application. It can operate offline after models are downloaded and can also use provider keys when a local model is too slow or incapable. The practical experience depends heavily on available RAM, GPU support, model format, quantization, and the quality of each model.
The core desktop application and local use are free and open source. Costs can include the user's hardware, storage, electricity, and any cloud model account connected to Jan; provider pricing and data terms apply directly. Roadmap items should not be treated as current features, so teams should evaluate the released desktop and CLI behavior they can verify today.
Local-first reduces routine data transfer but does not secure the whole workstation. Downloaded models and extensions require provenance checks, local chat files need device protection, verbose server logs can contain prompts, and MCP tools may reach files, browsers, or external services. Keep analytics disabled if desired, bind servers to localhost, use strong API keys, disable unneeded CORS and logs, and review every cloud or tool handoff.
How Jan works
Jan downloads compatible model files and runs them on the user's computer with local inference engines, storing chats and settings in its data folder. A user can chat in the desktop interface, attach permitted context, connect MCP tools, or start an authenticated local OpenAI-compatible server for other applications. Optional cloud providers receive prompts directly when selected. The model generates the response and may request tool calls; users remain responsible for model licenses, device access, provider terms, tool authorization, and reviewing outputs.
How to set up Jan
Confirm device fit
Check supported operating system, memory and acceleration, then choose a small model with an acceptable license, language, context, and safety profile.
Install and set privacy choices
Use the official download, review analytics consent and the data folder, encrypt the device, and keep the app updated.
Download one verified model
Confirm the publisher and model card, test basic chat offline, and avoid collecting large model files without storage and provenance controls.
Secure tools and API
Approve MCP servers individually, keep the API on 127.0.0.1, set a strong key, restrict trusted hosts and CORS, and disable sensitive verbose logs.
Evaluate before routine use
Test factuality, private documents, tool errors, prompt injection, speed and resource pressure; document when cloud use and human review are allowed.
Jan FAQs
Is Jan free?
Yes. The local application is open source and free. Hardware costs and any third-party cloud model usage remain the user's responsibility.
Can Jan work offline?
Yes, after the app and a compatible local model are downloaded. Cloud providers, online model search, updates, and remote MCP tools still need internet access.
Where are Jan chats stored?
Local workflows store data in Jan's data folder on the device. Connected cloud providers process the messages sent to their models under their own policies.
Does Jan provide an API?
Yes. Its local server exposes an OpenAI-compatible chat endpoint and requires an API key. Keep it bound to localhost unless the network is deliberately secured.
Are MCP tools safe to enable?
Not automatically. A tool may access files, accounts, browsers, or networks. Inspect its source and permissions, use least privilege, and confirm consequential calls.
Listing reviewed 2026-08-02. Product details and pricing can change; verify important terms on the provider's website.
Related Productivity AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.