---

4.6 Generative LLM keys ¶

A second, optional vendor seam, configured independently of the embedding keys above. It powers OwnSona’s generative background features — automatic consolidation, automatic conflict resolution, and relation-graph extraction (see Background intelligence). Leave LLM_API_KEY unset and the server runs purely on embeddings plus deterministic heuristics; every generative feature stays off and nothing breaks. Generative calls are background/maintenance only — they never sit on the synchronous recall / remember path.

You can point these at a different vendor (and a cheaper model) than your embeddings provider — any OpenAI-compatible /v1/chat/completions endpoint works (OpenAI, OpenRouter, a local Ollama, …).

LLM_API_KEY

Unset by default. When unset, the generative seam is not constructed and all generative features are disabled.

LLM_MODEL

Required iff LLM_API_KEY is set. The chat model id, e.g. gpt-4o-mini.

LLM_ENDPOINT

Required iff LLM_API_KEY is set. The OpenAI-compatible chat-completions URL, e.g. https://api.openai.com/v1/chat/completions.

LLM_PROVIDER

Default openai. A label recorded for logging/provenance only.