Docs · Platform
How Model choice works
How a request reaches a model.
1Check the budget
Deixic checks each call against organization, workspace, and team budgets and stops it if it would go over.
2Route to the chosen model
The request goes to your profile's model, on your provider keys.
3Fail over on a bad day
Authentication failures, rate limits, and server errors move the call to the next provider, other client errors stop, and a failing provider is paused.
4Record the cost
Usage lands in spend by agent, so every team sees what its agents cost.
Change models with company methods and checks intact
- Your keys, sealed
- Provider keys are encrypted with a cloud key management service, and the key service will not start in production without one; Pro and Max include model usage, and Enterprise runs on your keys by default.
- Connected means proven
- A model shows as connected only after a real call succeeds with your credential.
- Check a new model first
- Replay past work against a new model before you switch, and keep your context, skills, connections, and tests as they are.