AI coding agents have a chronic problem: every new session is like amnesia. The project standards you taught yesterday are forgotten today.
agentmemory solves this. It gained 8000+ stars in a week on GitHub, breaking 14k total — the fastest growing agent memory tool right now.
What It Does
Simply put, agentmemory adds a "long-term memory" layer to agents. It doesn't just store conversation history — that's been done before — it embeds a persistent knowledge management system into the agent's workflow.
Agents can save key info during sessions: project specs, common commands, pitfalls encountered, decision records. Next session, this info auto-loads without re-explanation.
Why It's Growing So Fast
This isn't surprising — it hits a real pain point. Anyone using Claude Code or Codex has experienced: you spend an hour telling the agent "our project uses X framework, Y conventions, Z agreements," and next day it asks from scratch again.
agentmemory turns "teaching the agent" from a one-time thing into an accumulation process.
It supports multiple agent platforms: Claude Code (.claude-plugin), Codex (.codex-plugin), and via MCP protocol, any MCP-supporting agent.
Technical Details
TypeScript, 374 commits, latest v0.9.21 released 9 hours ago. Very active.
Core components: memory store (local filesystem + remote deploy), benchmark tools (load-100k harness with p50/p90/p99), plugin system.
Main sources: