widemem.ai
JSON →Open-source AI memory layer with importance scoring, temporal decay, hierarchical memory, and YMYL prioritization
Install
pip install widemem-ai[faiss] Tools · 7
- batch_conflict_resolution Single LLM call for all facts vs. existing memories, reducing N facts to 1 API call.
- importance_and_decay Facts rated 1-10 with exponential/linear/step decay, ensuring old trivia fades while critical facts persist.
- hierarchical_memory Facts roll up into summaries and themes, auto-routed for broad or specific queries.
- active_retrieval Contradiction detection plus clarifying questions to resolve conflicting memories.
- ymyl_prioritization Health, legal, and financial facts receive higher importance floors, immunity from decay, and forced contradiction detection.
- confidence_and_abstention Returns confidence level (HIGH/MODERATE/LOW/NONE) for every retrieval, allowing agents to abstain instead of guessing.
- retrieval_modes Three modes (strict, helpful, creative) for handling low-confidence memory misses, enabling appropriate agent behavior.
Environment variables
QDRANT_URL
Links
★ 45 GitHub stars