Engram

Engram provides a novel approach to persistent information management for agents, moving beyond simple data storage to create a dynamic understanding of past interactions. Key features include: • Knowledge Graph for connected memories • Sleep Cycles for knowledge consolidation • Spreading Activation for contextual recall • Entity tracking and contradiction detection • TypeScript SDK and REST API for flexible integration This system addresses the challenge of agents "forgetting" past preferences and contexts by automatically extracting, connecting, and surfacing relevant information. Instead of relying on expensive full conversation histories or tedious manual file management, Engram distills hundreds of raw episodes into structured knowledge. This process is akin to how a human brain consolidates memories during sleep, ensuring that every new session doesn't start from zero. The system utilizes significantly fewer tokens compared to full context dumping, making it highly efficient. It builds a web of understanding where memories connect via typed edges (e.g., supports, contradicts, elaborates), allowing for rich, contextual recall. For example, querying "Thomas" could cascade to recall related facts like "marathon" and "prefers morning runs." Engram is ideal for developers building sophisticated agents that require deep, persistent understanding of user preferences and project details. It integrates seamlessly with platforms like Claude Code and Cursor, providing agents with ten crucial memory tools out-of-the-box. The open-source TypeScript SDK with zero dependencies ensures easy setup and customization for various agent-based applications.