Show HN: Engram – Persistent memory for AI coding agents (open source) (github.com)

A new open-source project called Engram has been launched, providing a universal memory layer for AI coding agents. Unlike traditional memory systems that rely on flat vectors or manual curation, Engram uses knowledge graphs with typed edges, allowing for advanced features like sleep-cycle consolidation and context-aware retrieval. The tool boasts improved performance benchmarks, achieving 79.6% accuracy in memory recall with 44% fewer tokens used per query compared to manual memory files, indicating a significant enhancement in efficiency and effectiveness for AI agents handling information. Engram introduces ten different memory tools through the MCP (Memory Consolidation Protocol), enabling functionalities such as auto-extraction of entities, proactive memory surfacing, and semantic search for recall. Key features include the ability to create relationships between memories and conduct automated conversations through structured memory ingestion. The project aims to enhance the AI/ML community's approach to persistent memory, shifting focus from data storage to a deeper understanding and contextual awareness, which can lead to more intelligent and capable AI agents.