Can I bring my own embedder for dense retrieval?

Yes. Dense retrieval in the daemon is bring-your-own-embedder by design. Point it at whatever serves embeddings in your stack: Ollama, vLLM, TEI (Text Embeddings Inference), llama.cpp, or LiteLLM as a router in front of anything else.

How it composes:

• The dense lane fuses with the built-in BM25-class lexical lane and graph signals; you get hybrid retrieval, not a vector-only replacement.
• The local dense lane is never metered. Your embedder, your hardware, your cost model; CueCrux charges nothing for it.
• Without an embedder configured, lexical plus graph carries the load. Dense is an upgrade path, not a prerequisite.

Hosted options exist where they add something local cannot: better-dense rerank comes with Pro, and the richer entity, event, and navigational lanes run on hosted CoreCrux. Both are optional and both meter transparently, with a signed receipt per spend.