Embeddings

Enable the optional vector index for semantic and hybrid search.

[embeddings]
enabled = true
provider = "ollama"           # "ollama" | "openai" | "mock"; falls back to [llm] provider
model = "nomic-embed-text"    # provider default if unset (openai: text-embedding-3-small)
api-key-env = "OPENAI_API_KEY"
cache = true                  # disk cache in .ekos/embed-cache/

Vectors are built by ekos commit after the other post-commit steps. Then:

ekos query find "customer churn" --mode vector
ekos query find "customer churn" --mode hybrid

Hybrid mode fuses BM25 and cosine rankings with reciprocal-rank fusion. The vector index is a brute-force cosine index — appropriate to the size of a typical compiled knowledge base, not billions of vectors.

Optional rerank

[retrieval]
rerank = "llm"
rerank-candidates = 20

Off by default so ranking is reproducible.