Decision API: classify and score items with probabilities

Many automations hinge on a judgment call: which team should get this ticket, is this lead a good fit, does this email need a reply. The new decision service answers questions like these with a probability for every possible answer, so your workflow can act only when it's confident enough. Like every built-in service, it needs no API key of your own and bills to your workspace.

How it works

• Three kinds of question. A choice among up to 255 options, a score on up to 10 ordered levels, or a yes or no. One request can ask several questions about the same item.
• Any item. Send text or JSON, up to 32,000 input tokens.
• Thresholds you control. A new decisions skill teaches specialists how to phrase questions and set a confidence threshold for each action.
• Stable models. A request always names its model and is never silently moved to a newer one, because thresholds are tuned per model. At launch the model is TypeSafe's Jev, and you can bring your own TypeSafe key.
• Clear failures. Errors come back as one of six codes, and a provider's raw error never reaches you.
• Proven in-house. Our own Auto model routing and New task routing run on the same service.

Also shipped

• Calls to our chat completion service can now name an exact model, such as Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, or Gemini 3.8 Flash.