Add a cloud model per thread with your own key
Use one app for both local and cloud models
Jan is offline-first, but it is not local-only. You can connect a cloud provider — OpenAI (GPT), Anthropic (Claude), Mistral, or Groq, among others — by adding your own API key, and then choose a cloud model on a per-thread basis. This means one familiar app for everything: a local model for sensitive work, and a frontier cloud model for the occasional hard question, without switching tools. The cloud models cost whatever that provider charges; local models stay free.
- 1 Get an API key from a provider you already use (OpenAI, Anthropic, Mistral, or Groq). Copy the key.
- 2 Open Jan's settings and find the provider/model-providers section. Paste your key for that provider and save.
- 3 Start a new thread and use the model picker to choose one of that provider's cloud models instead of a local one.
- 4 Send a hard question that benefits from a frontier model — for example:
Critique the experimental design in this abstract and suggest two controls I may be missing. - 5 Switch back to a local model for your next sensitive thread. Notice that the choice is per thread: cloud when you want capability, local when you want privacy.
You ran one thread on a cloud model (using your own key) and another on a local model, in the same Jan app, choosing the model per thread.
Write your own local-vs-cloud rule
The value of having both in one app is knowing when to use each. Make that decision explicit for your own work.
Run the same research question once on a local model and once on a cloud model, then write a short rule for when each is the right choice.
- 1 Pick a real question from your current work.
- 2 Ask it in a local-model thread, then ask the identical question in a cloud-model thread.
- 3 Compare answer quality, speed, and — crucially — what left your machine in each case.
- 4 Write a one-sentence rule: 'I will use local Jan for X and a cloud model for Y.'
Both answers and your one-sentence local-vs-cloud rule.