Tensory

•
4
•

Provides zero-setup, claim-native, persistent memory for AI agents with advanced cognitive mechanisms and collision detection, all within a single SQLite file.

Tensory offers an embedded, claim-native memory solution specifically designed for AI agents, featuring a single SQLite file backend and built-in collision detection. It transforms raw text into atomic, verifiable claims, applying context-aware extraction and advanced cognitive mechanisms like salience decay and surprise scoring without relying on additional LLM calls. This tool eliminates the need for complex infrastructure, offering a robust, persistent memory store for large language models and other AI clients, ideal for enhancing long-term conversational memory and knowledge management.

主要功能

01Hybrid search (FTS5 + vector + graph traversal) for retrieval
02Web dashboard for entity graph exploration and memory stats
03Cognitive mechanisms: salience decay, surprise scoring, priming
044 GitHub stars
05Claim-native storage for atomic, verifiable statements
06Built-in collision and contradiction detection for facts

使用案例

01Integrating persistent memory into AI agents like Claude and Cursor
02Enhancing long-term conversational memory for LLM applications
03Building advanced cognitive architectures for AI assistants and research agents