Agentmemory

Agentmemory is a persistent memory system designed specifically for AI coding agents. It acts as a memory layer that automatically captures everything an AI agent does during a coding session — every tool call, every prompt, every stop command — and stores it for future recall. The system compresses these observations into semantic memories that can be retrieved in milliseconds, giving AI agents the ability to remember past work and build upon it.

The tool works through three core primitives: hooks that automatically capture agent activities, a recall system that uses hybrid retrieval (BM25, vector search, and knowledge graph), and a consolidation engine that periodically compresses raw data into meaningful memories. What makes Agentmemory stand out is its zero-external-database architecture — it runs entirely as a single Node.js process without requiring Redis, Postgres, or other external systems. It includes a real-time viewer on port 3113 for monitoring memory activity and supports multiple AI providers including Claude, Gemini, and OpenRouter.

This tool benefits developers building or using AI coding assistants who want those assistants to maintain context across sessions. A developer working with Claude Code could use Agentmemory to ensure their AI remembers architectural decisions from previous sessions. Teams building custom coding agents can integrate it to give their agents persistent memory capabilities without managing complex database infrastructure. Researchers studying agent behavior can use it to analyze and replay complete session histories for debugging and improvement.