qavr-memory

Rank memories by usefulness by applying learned Q-value scoring to re-rank vector-store results, prioritizing memories that previously led to successful outcomes. This enables warm-context prioritization for agents handling long-running conversations.

Q-value scoring improves retrieval relevance by storing per-memory utility values and updating them based on task outcomes, ensuring frequently useful memories are promoted over merely semantically similar ones.

Yes, this memory ranking approach is designed for long-running conversation histories and knowledge bases, applying cold-to-warm context adaptation that becomes active after sufficient interactions to re-rank vector-store results.

Prioritize frequently useful memories in a Python knowledge base by integrating Q-value scoring with your vector store and running a query like qavr_query to obtain prioritized results based on actual usefulness.

Vector stores fail to promote useful memories because they rely solely on semantic similarity. Adding learned utility scoring updates per-memory Q-values on outcomes to re-rank results and improve retrieval relevance over time.

Avoid learned utility scoring for memory retrieval during cold-start periods with insufficient interactions, as the Q-value ranking requires outcome data to adapt from cold to warm context and effectively prioritize memories.