API reference

Every ingress request needs an API key as a Bearer token (Anthropic clients may also use x-api-key):

Authorization: Bearer llm11_live_...

Keys prefixed llm11_sandbox_ behave identically but are labelled sandbox on every receipt, for testing against real models without touching a production credit balance’s optics.

OpenAI-compatible. Accepts the standard messages, temperature, max_tokens and response_format, plus one addition:

{
  "messages": [...],
  "context": "optional source text the groundedness check verifies claims against"
}

Returns a standard chat completion plus _llm11: the full receipt (model, baseline, verdict, every rung run, cost, savings).

Anthropic Messages-API-compatible. Same engine, same receipt, translated at the edges - point an Anthropic SDK’s base URL here and it works unmodified.

Structured, with a stable type:

{ "error": { "message": "...", "type": "insufficient_credits" } }

• 401 invalid_request_error - missing or invalid API key
• 402 insufficient_credits - prepaid balance is $0.00
• 502 upstream_error - the model provider failed
• 501 not_implemented - streaming (not yet supported on this deployment)

No hard per-key rate limit beyond your plan’s monthly verified-request volume today; a request past that volume is billed at your plan’s overage rate rather than rejected. Streaming responses are not yet implemented - a stream: true request receives a clear 501 rather than a silent non-streamed response pretending to be one.