# LLM extraction

LLM extraction is BYOK. Launch providers are OpenAI, Anthropic, Gemini, and OpenRouter; arbitrary base URLs are not accepted. `POST /v1/extract` fetches one source through the same caller-owned proxy path as scraping and returns source URL, final URL, byte count, SHA-256 evidence, and advisory schema-validated output. The user supplies an encrypted saved connector or a request-scoped key and must explicitly consent to third-party data transfer. Input is capped at 100,000 bytes and output at 2,000 tokens. The `llm:use` scope does not override tier limits. Prompts, page bodies, credentials, target authorization, and raw provider responses are excluded from telemetry and public output.