AgentMemory is an open-source persistent-memory system for coding agents and other MCP-capable assistants. It is designed to stop users from repeatedly explaining the same architecture, decisions, bugs, and preferences by capturing useful information from agent sessions and making it available again when relevant.

The system combines lifecycle management, confidence scoring, knowledge graphs, and hybrid search and can be shared across multiple agent clients. It supports environments including Claude Code, GitHub Copilot CLI, Cursor, Gemini CLI, Codex CLI, OpenCode, and other tools that can connect through hooks, MCP, or HTTP.

AgentMemory is useful for developers who work with AI assistants across long-running projects and multiple sessions. Instead of relying only on small static instruction files, it creates a searchable memory service that can accumulate project knowledge over time and inject appropriate context into future work across different agents.