Agentmemory

Agentmemory provides persistent memory for coding agents, enabling them to retain and recall information across sessions with high efficiency. Key features include: * Infinite memory for coding agents * Significantly fewer tokens per session * Up to 200x more tool calls without context limits * 100% open-source and runs locally * No external databases required This runtime addresses the challenge of context window limitations by employing a triple-stream retrieval mechanism: BM25, vector, and knowledge graph. This hybrid approach ensures comprehensive recall, reranked on device for speed, with a P50 under 20ms on a laptop. Hourly sweeps consolidate raw observations into semantic memories, merging duplicates and decaying stale rows to maintain optimal performance and relevance. Built on the III engine, Agentmemory operates as a complete memory runtime, encompassing capture, recall, consolidate, observe, and federate functionalities. It includes 12 auto-capture hooks that pipe every tool call, prompt, and stop into the memory pipeline, compressing observations. Additionally, it offers 51 MCP tools and 121 REST endpoints, providing a full surface for memory management, including save, recall, smart search, session management, governance, audit, and export functions. Agentmemory is ideal for developers and engineering teams working with coding assistants that require extended context retention and efficient information retrieval across numerous coding sessions. It enhances productivity by allowing agents to remember past interactions and learn from them without hitting typical context limitations, ensuring a seamless and continuous development experience.