HomeToolsMCPHow It WorksStoriesPhilosophyCommunityArchitectureStar on GitHub
All Tools
A
Dev ToolsFreeOpen Source

AGENT MEMORY

Persistent memory layer for AI coding agents

Apache-2.0

ABOUT

AI coding agents have no persistent memory across sessions, forcing developers to waste the first 5 minutes of every session re-explaining their stack, architecture decisions, preferences, and bug context. Built-in agent memory (CLAUDE.md, .cursorrules) caps out at ~200 lines and goes stale. Agent Memory silently captures every tool use, compresses it into structured 4-tier memory (working, episodic, semantic, procedural), and injects the right context at the next session start -- eliminating re-explanation entirely.

INSTALL
npm install -g @agentmemory/agentmemory npx @agentmemory/agentmemory

INTEGRATION GUIDE

1. Persistent cross-session memory for coding agents -- agents remember architecture decisions, code patterns, preferences, and bug fixes from previous sessions without re-explanation 2. Multi-agent memory coordination -- share memory across Claude Code, Codex CLI, Cursor, Gemini CLI, OpenCode, and any MCP-compatible agent with leases, signals, and team namespaces 3. Session replay and debugging -- replay any captured session via the real-time viewer on :3113, scrubbing through prompts, tool calls, and responses as discrete events 4. Session JSONL import -- import existing Claude Code JSONL transcripts to bring older sessions into the memory runtime for search and replay

TAGS

agent-memorypersistent-memorymcpcoding-agentknowledge-graphbm25vector-searchcross-sessiondeveloper-tools