AI Providers and Routing
Configure multiple AI providers and choose which model handles each feature.
At a glance
- Add provider — "+ Add provider" button to choose from the 21 supported ones or Ollama
- Save the key — encrypted with the OS-level safeStorage (never transmitted off your computer)
- Test — verify the key + model actually work with a real ping
- Load Models — calls the provider's /v1/models, then probes each with a 1-token request to verify what your API key can call (badges "ok" / "no access" / "error")
- Tick the models to enable — only these appear in the routing dropdowns
- Anthropic — does not expose /v1/models, so you enter model IDs manually (e.g. claude-sonnet-4-6)
- Routing — assign each feature (Default, Inbox replies, Comments, Messages, Posts, Purpose) a provider/model pair
- Default is the fallback used when a specific feature has no route set
How it works
If you use Booost AI there is nothing to set up here: routes are for people who bring their own keys.
With your own keys, every app feature asks for its route; if that feature has none, it uses the Default; if that is missing too, generation stops with an error saying so. Which is why it pays to set the Default before anything else.
Each provider stays on its own: separate key, separate enabled models. That lets you send bulk work to a fast, cheap model and keep the good one for the copy that matters.