An LLM router built to control API cost

How automatic routing works

1. Your application sends a text messages request.
2. The router assesses the task and supported model choices.
3. It selects a cost-effective model intended to handle that task.
4. Basktre forwards the request without retaining prompt or response bodies.
5. The response identifies the model used and provides usage metadata.

{
  "model": "auto",
  "messages": [
    {
      "role": "user",
      "content": "Summarize this support ticket in three bullets."
    }
  ]
}

When automatic routing is appropriate

• High-volume text workloads where several models can meet the requirement.
• Summarization, extraction and routine generation where cost flexibility matters.
• Applications that record the selected model and evaluate outputs continuously.

When to pin a model

• You require repeatable model-specific behavior or have validated only one model.
• Your prompt relies on a capability or parameter that is not shared by every model.
• You are running benchmarks, regulated review workflows or incident reproduction.

Current limitations

Automatic selection is not a guarantee of a particular quality, latency or saving. Basktre’s public documentation does not currently promise a configurable fallback chain, voice routing, image generation, JSON mode or tool-call normalization. Use a pinned model and test every required parameter when those details matter.

How to evaluate savings responsibly

Basktre has not published a sufficiently detailed benchmark to support a universal percentage-savings claim, so this page does not make one. Build a representative prompt set, define pass/fail quality criteria, record the chosen model and token cost, and compare the same successful requests with your pinned baseline. Include Basktre’s 4% wallet top-up fee in the final calculation.

Compare current rates on the LLM API pricing page or model a workload with the cost calculator.