# llm11

> A verification layer for LLM calls. Catch unsupported answers before your users see them, route the cheap ones to a cheaper model, and get a receipt for both.

llm11 sits in front of any OpenAI- or Anthropic-compatible model call. It triages each
request (Jev when configured, a heuristic backend otherwise), routes it to the cheapest model
likely to answer it well, checks the answer against a verification ladder scaled to how much the
call actually needs, and escalates to a stronger model when a check fails. Every request returns a
receipt: which checks ran, what each one found, and the real dollar cost against your own declared
baseline model.

## Integrate

- OpenAI-compatible: POST https://llm11.com/v1/chat/completions with an llm11_live_… API key as a Bearer token.
- Anthropic-compatible: POST https://llm11.com/v1/messages (Bearer token or x-api-key header).
- MCP: POST https://llm11.com/api/mcp (JSON-RPC 2.0), tool `verify_request`.
- OpenAPI: https://llm11.com/api/v1/openapi.json
- Machine-readable pricing: https://llm11.com/pricing.json

## What we detect

Claims unsupported by source text you supply; a second independent model materially disagreeing;
schema, format and reference violations; and an unstable answer under resampling. We do not detect
free-floating factual error with no source to check against, and we do not claim to eliminate
hallucination.

## Docs

- https://llm11.com/docs/quickstart
- https://llm11.com/docs/api
- https://llm11.com/pricing