Mengram

Mengram is an AI memory layer that stores facts, events, and workflows in a shared graph, solving the problem of AI agents losing context between sessions. It targets developers and engineering teams who use AI coding tools like Claude, Cursor, and Codex, as well as builders of agent products. It is positioned as an open-source, self-hostable memory backend with a one-click connector for Claude and an MCP server, offering low-latency memory extraction and retrieval.

Key features

• One-click Claude connector
• MCP server with 30 memory tools
• Three memory types: semantic, episodic, procedural
• Survives /clear and auto-compaction
• Isolated memory per user_id
• REST + MCP + webhooks
• Server-side extraction
• Open source Apache-2.0
• Self-hostable
• No vendor lock-in
• Less than 50ms latency