Embedders¶

Embedders turn text into vectors. They are only needed by retrievers that embed locally (memory, faiss, qdrant, pinecone, pgvector, and elasticsearch in knn mode). Server-side embedding backends (chroma, weaviate) ignore this setting and embed internally.

You configure an embedder once under retriever.embedder and OpenAgent Eval injects it automatically into the retriever.

config.yaml

retriever:
  provider: memory
  embedder:
    provider: sentence_transformers
    model: all-MiniLM-L6-v2

Comparison matrix¶

Embedder	Install extra	Local?	Default model	Needs key?
Sentence-Transformers	`openagent-eval[evaluation]`	✅	`all-MiniLM-L6-v2` (384-dim)	❌
Mock	(built-in)	✅	deterministic hash vectors	❌

Which one should a beginner pick?¶

Real semantic search? Use sentence_transformers with all-MiniLM-L6-v2 — small, fast, runs on CPU, no API key.
Offline tests / CI? Use mock — deterministic vectors, nothing to install.

Common configuration¶

Setting	Type	Default	Notes
`provider`	`str`	—	Required. `sentence_transformers` or `mock`.
`model`	`str`	`all-MiniLM-L6-v2`	Model id (sentence-transformers) or ignored (mock).
`settings.device`	`str \\| null`	`null`	`cpu` / `cuda` (sentence-transformers only).
`settings.dimension`	`int`	`32`	Vector size (mock only).

Next steps¶

Open an embedder page above for a full walkthrough.
See which retrievers need an embedder in ../retrievers/index.md.