Deciding which AI to use, automated.
Give it required capabilities and a minimum quality tier. It picks the cheapest registered model that qualifies, with the reasoning spelled out. Something else actually calls the model.
Screens
What it does
- 01Deterministic cost-optimal routing
- Filter by capability, then quality tier, then pick the cheapest of what remains. Same inputs, same result, every time.
- 02A model registry
- Claude, GPT, and Gemini models ship pre-seeded with rough pricing. Edit them to match your contract, or add your own.
- 03Verdicts with reasoning
- See exactly why a model was chosen — or exactly why nothing qualified.
- 04Routing history
- Every decision is kept, so you can see which models get picked most often.
- 05A CLI for CI/CD
- The bundled trcli prints the chosen model to stdout and exits 1 when nothing matches.
- 06It never calls an AI provider
- It only decides. Sending the request to the chosen model is up to your code.
How it works
Review or register models
Edit the pre-seeded pricing to match your contract, or add your own models.
Request a route
Pass a task, required capabilities, and a minimum quality tier — manually or via trcli from CI.
Use the result
Your own system sends the actual request to whichever model came back.
Good to know
- Deterministic verdicts — the same constraints and registry always produce the same result.
- No match still comes with a reason — never a silent null.
- It never calls an AI provider — no API keys or provider integration to manage.