Claude-mem

Claude-mem and cmem provide an open-source memory layer for AI agents, enabling persistent, structured recall of agent actions and decisions across sessions, machines, and development environments. This system addresses the challenge of AI agents losing context between sessions by capturing every decision, bugfix, and dead end as structured observations in a temporal database, allowing agents to resume work with full historical context regardless of where or when they last operated.

The core offering consists of the claude-mem open-source engine, which can be installed with a single command and requires no configuration or account to begin. This engine writes observations automatically as agents work, storing them in a database that supports both full-text and recency-based hybrid search. The system is designed for compatibility with a wide range of agent frameworks, IDEs, and Model Context Protocol (MCP) clients, including named integrations such as Claude Code, Cursor, Windsurf, OpenCode, Codex CLI, and Gemini CLI. Developers can run claude-mem locally or use CMEM Cloud, which provides a cloud sync layer that mirrors the local database and offers a private MCP endpoint for access from any machine or agent.

CMEM Cloud is available in early access and is free during this period. 0 license. The platform emphasizes privacy and flexibility, supporting offline-first operation and end-to-end private access via personal MCP links. Teams can share a single memory across multiple agents and editors, facilitating collaborative development and long-term traceability of project decisions and changes. The system's hybrid search enables agents to recall relevant information in milliseconds, even from projects that have not been touched in months.

By integrating directly with agent frameworks and development tools, claude-mem and cmem streamline the process of building and maintaining persistent AI agent memory, reducing the need for manual context management and enabling agents to maintain continuity across diverse workflows.