Jan is the project behind this setup. Jan is a desktop application, so this guide does not require Docker, a domain or a server. Model size is the main decision.
- Recommended method
- Desktop application
- Difficulty
- Beginner
- Typical time
- 15–45 minutes plus download
What you need
- Windows 10+, a supported Linux distribution or Apple Silicon Mac
- At least 8 GB RAM; more is better
- Enough disk space for model files
- Current graphics drivers if using GPU acceleration
Set it up
Install the official desktop release
Use Jan's current installer for the operating system. Linux users should prefer the documented Flatpak route when available.
Let the default model finish
Jan's quick start downloads a default model. Test it before browsing larger models.
Check hardware acceleration
Open Settings and System Monitor. Confirm the detected GPU or accept CPU operation if the machine has no supported accelerator.
Try a private offline prompt
Disconnect from the network after the model is downloaded and confirm a normal chat still works.
Locate and back up the data folder
Jan documents where threads, models and settings live. Back up only the material you need; model files can usually be downloaded again.
Reserved for launch
Verify it works
Do not call the setup finished until these small tests pass:
- A local model answers with the network disconnected
- System Monitor shows the expected hardware
- Memory use remains stable
- The data folder location is known
Protect the service
- Cloud providers are not local merely because they appear inside Jan
- Do not paste secrets into untrusted downloaded models or extensions
- Keep the app and inference backend updated
- Review logs before sharing them publicly
Official resources
Videos are supplemental. Use the official documentation for current commands, versions and security guidance.