Cross-encoder reranker

Reorders retrieved memories with a local sentence-transformers cross-encoder model, selected via RERANKER_PROVIDER=cross_encoder. Reranking reorders retrieved memories by relevance after hybrid search.

Enable it

RERANKER_PROVIDER=cross_encoder
CROSS_ENCODER_MODEL=cross-encoder/ms-marco-MiniLM-L-6-v2

Requires sentence-transformers: pip install sentence-transformers.

Configuration

Variable	Default	Description
RERANKER_PROVIDER	none	Set to cross_encoder.
CROSS_ENCODER_MODEL	cross-encoder/ms-marco-MiniLM-L-6-v2	Cross-encoder model to load.

Notes

• Runs entirely locally; no API key or network call is needed at query time.
• The model is loaded lazily on the first rerank call.
• The ms-marco model emits raw logits, which are sigmoid-mapped into [0, 1] so downstream scores stay on a stable scale.