Skip to content
FitRank

Platform and integrations · Decision API

Ask typed AI questions from your own tools

The typed-decision engine behind matching is open to your other systems. Each question has a fixed list of options and declared inputs; you get a probability for every option, or an abstain when the model isn't sure.

POST /api/v1/decide
Authorization: Bearer <project key>

{
  "decision": "agent.model_choice",
  "inputs": {
    "task": "Rename a variable across one file",
    "context": "Small change, no earlier attempts"
  }
}

→ 200 OK
{
  "decision": "agent.model_choice",
  "answer": "cheap",
  "probabilities": { "cheap": 0.93, "strong": 0.07 },
  "confidence": 0.93,
  "abstained": false,
  "flags": [],
  ...
}
Illustrative call to a built-in question: a probability for every option, or an abstain when the model isn't sure.

The problem

Calling a chatbot from internal tools returns free text you can't test, cap or audit, and costs grow with every call.

How it works

  1. 1Create a project and an API key in the admin area.
  2. 2Call a question by name with its inputs, from your service or the Python client. Built-in questions help AI agents choose a cheap or strong model, approve a tool action and check a result.
  3. 3Get a probability for every option, or an abstain when the model isn't sure. Inputs that state protected attributes are refused.
  4. 4Every call is logged, rate-limited and counted against the project's monthly budget.

What's included

Where teams use it

Questions about decision api

Can the API return free text?

No. It returns a probability for each of the question's options, or an abstain. That's what makes the answers testable.

Which plan includes the Decision API?

Enterprise, with per-project keys and budgets.

More of FitRank

See all 19 capabilities

Be one of the first teams to staff with FitRank

We're taking on a small number of pilot companies. Pre-register for early access, or talk to us about an annual licence.