LLM reranker

Reorders retrieved memories by asking the app's configured LLM to score each document, selected via RERANKER_PROVIDER=llm. Reranking reorders retrieved memories by relevance after hybrid search.

Enable it

RERANKER_PROVIDER=llm
LLM_PROVIDER=<your-llm-provider>

Configuration

Variable	Default	Description
RERANKER_PROVIDER	none	Set to llm.
LLM_PROVIDER	—	LLM backend used to score documents; built via the LLM factory.

Notes

• Uses the same LLM configured by LLM_PROVIDER, built through the LLM factory.
• A single call scores all documents from 0 to 10 to keep token cost low.
• If scoring fails, it degrades gracefully by preserving input order with neutral scores.