TL;DR
TencentDB Agent Memory is a fully local memory tool that enables AI agents to hierarchically structure and retrieve conversation history, operating without external APIs and significantly reducing token costs.
Key features
Symbolic short-term memory: Compresses heavy tool logs into concise Mermaid symbols, reducing token usage by up to 61%.
Hierarchical long-term memory: Distills fragmented conversations into structured personas and scenes, storing them in meaningful hierarchies instead of flat vector stores.
4-tier progressive pipeline: Gradual transition from short-term to long-term memory improves context retention and retrieval accuracy.
Fully local: Operates entirely locally with no external API dependencies, ensuring privacy.
When to use it
When long-running agents need to remember recurring contexts (e.g., project background, SOPs, tool rules).
When aiming to reduce token costs while improving agent task success rates.
When a fully local memory system is required without external API or cloud dependencies.