Can I use this fully offline with no OpenAI API key?

[unusdon]

Yes — point both providers at Ollama. Install Ollama, then:

ollama serve
ollama pull llama3.2 nomic-embed-text

In .env.local:

CHAT_PROVIDER=ollama
EMBEDDING_PROVIDER=ollama

Restart npm run dev and npm run worker. The chat path, retrieval, and ingestion all run locally with zero API spend.

[unusdon]

Yes — this works end-to-end. One thing to flag: nomic-embed-text is 768-dim; the chunk column is vector(1536), so the embed module zero-pads to 1536 before insert. Cosine similarity in the padded space is mathematically equivalent to similarity in the 768-d subspace, so retrieval quality is preserved. The llama3.2 chat model handles citations well in our tests — make sure ollama pull succeeded for both models before starting the worker.