Skip to main content

LLM Gateway

A thin gateway in front of upstream LLM providers (OpenAI, Anthropic, local Ollama, etc.). Every completion request is wrapped in a CKO so the response carries an operator chain, zeqond tick, and ZeqProof — meaning the model output itself becomes a verifiable computation in the ZeqState log.

GET /api/llm/providers​

Public. Lists configured providers and which models they expose.

{
  "providers": [
    { "id": "openai", "models": ["gpt-4o","gpt-4o-mini"] },
    { "id": "anthropic", "models": ["claude-opus-4-6","claude-sonnet-4-6"] }
  ]
}

GET /api/llm/models​

Flat list of all model IDs across providers, with context window and pricing where available.

POST /api/llm/key​

Stores a user-supplied upstream key, encrypted with the user's ZID-derived field key.

{ "provider": "openai", "key": "sk-..." }

POST /api/llm/chat​

Auth: API key. Rate limit: llmLimiter.

{
  "provider": "anthropic",
  "model": "claude-opus-4-6",
  "messages": [{ "role": "user", "content": "Hello" }]
}

Response:

{
  "completion": "Hi!",
  "provider": "anthropic",
  "model": "claude-opus-4-6",
  "operatorChain": ["KO42","ON0"],
  "zeqond": 65392856,
  "phase": 0.4127,
  "zeqProof": { "alg": "HMAC-SHA256", "value": "..." },
  "usage": { "promptTokens": 5, "completionTokens": 3 }
}

Every chat completion publishes a free-tier row to ZeqState (endpoint: /api/llm/chat).