Memory LanceDB

Name: Memory LanceDB
Author: cablate

bycablate

0•

API Development

Developer Tools

Data Science & ML

Equips AI agents with persistent, intelligent long-term memory via hybrid retrieval, cross-encoder reranking, and dynamic lifecycle management.

AI agents often suffer from short-term memory, forgetting context between sessions. This tool solves that by providing a robust, LanceDB-backed long-term memory system. It integrates advanced retrieval methods like vector and BM25 search, enhanced by cross-encoder reranking, to ensure the most relevant information is always surfaced. Key features include intelligent memory extraction, a dynamic lifecycle management system that promotes important memories while allowing stale ones to fade, and multi-scope isolation for managing separate memory contexts for agents, users, or projects.

Key Features

01Dynamic Memory Lifecycle (Weibull decay + 3-tier promotion)

02Smart Metadata Extraction and Noise Filtering for enhanced memory quality

03Hybrid Retrieval (Vector + BM25) with cross-encoder reranking

04Multi-Scope Isolation for per-agent, per-user, or per-project memory

05Compatibility with various embedding and rerank providers (OpenAI, Jina, Ollama, etc.)

060 GitHub stars

Use Cases

01Implementing self-improvement and learning capabilities for AI systems

02Managing isolated knowledge bases for different users, agents, or projects within an AI application

03Providing persistent memory for AI agents across multiple sessions