Engram Documentation

Engram is a toolkit designed to provide persistent memory and context management capabilities for AI agents. It offers a suite of tools including a TypeScript SDK, a REST API, Model Context Protocol (MCP) tools, and a command-line interface (CLI), enabling developers to integrate memory functions into their AI workflows. The platform allows agents to store, recall, and manage memories, supporting use cases that require long-term context retention and knowledge consolidation.

The toolkit features commands and API endpoints for storing memories with automatic entity and topic extraction, retrieving relevant memories using semantic search and spreading activation, generating structured context summaries, and running consolidation processes that distill episodic information into semantic knowledge. Additional capabilities include proactive surfacing of context, connecting memories with typed relationships, soft and hard deletion of memories, listing tracked entities, viewing vault statistics, and ingesting text for memory extraction. The MCP tools become available after initialization and support integration with environments such as Claude Code, Cursor, and other MCP-compatible clients. The init process detects the user's editor and can prompt for a Gemini API key to configure the environment and create a memory vault.

Engram can be installed via npm or Homebrew, and its REST API enables integration with agents written in Python or other languages. js agents, providing zero-network-overhead access to memory functions. The CLI and API offer endpoints for health checks, statistics, and entity management, supporting both single-language and multi-language agent setups.

The platform is positioned as a framework and set of tools for developers building AI agents that require persistent, structured memory and advanced context management.