Documentation
¶
Overview ¶
Command embed-server-quickstart wires a local OpenAI-compatible embedding server into the reliquary facade.
Tier 0 of the Embed→Judge stack: a quantized local embedding model (default nomicai-modernbert-embed-base-4bit) served behind an OpenAI-compatible /v1/embeddings endpoint embeds the corpus and every query at near-zero marginal cost. Nomic-family models require the `search_document: ` and `search_query: ` task prefixes, applied by the adapter from the request kind; the server honors the `dimensions` field, so Matryoshka truncation (for example a second 64-d embedder for approximate candidate search) is a second constructor with a different Dimensions value.
The caller owns the server process, endpoint, and credentials. Start your local embeddings server first (defaults below target http://localhost:8000/v1), then run from the repo root:
GOWORK=off go run ./examples/embed-server-quickstart
Override with EMBED_SERVER_URL, EMBED_SERVER_API_KEY, EMBED_MODEL, and EMBED_DIM.
Source Files