Agentmemory

Agentmemory is the memory layer your coding agent should have had from day one—and now it does. Built as a complete memory runtime for Claude Code, Codex, and every AI coding agent out there, Agentmemory captures every session, recalls context in milliseconds, and runs entirely on-device with zero external databases. No Redis. No Kafka. No Postgres. Just one process, 20.4K GitHub stars, and a runtime that actually works.

Here's what makes it different: 12 auto-capture hooks pipe every tool call, prompt, and stop directly into the memory pipeline without you writing a single line of glue code. Install the plugin, done. Their triple-stream retrieval combines BM25, vector, and knowledge graph scoring through an on-device reranker, achieving 95.2% R@5 on LongMemEval-S—that's 92% fewer input tokens per session while being more accurate. They're P50 under 20ms on a laptop. That's not a server farm. That's your local machine.

They've got 53 MCP toolsnative—memory_save, memory_recall, memory_smart_search, memory_sessions, governance, audit, export—the whole surface behind a single MCP server. Plus 125 REST endpoints if you want to curl it or proxy it from your own agent. Hourly sweeps automatically consolidate raw observations into semantic memories, merge duplicates, decay stale rows, and emit batched audit rows. 1312 tests passing. And they're Linux Foundation backed, ranked #19 in 2026 Live.

Everyone building AI agents right now is solving the same problem in their own messy way—trying to remember what happened five prompts ago. Agentmemory just solved it properly. If you're building with Claude Code or Codex, you need this. Your agent already forgets everything the moment you restart the session. Stop rebuilding memory from scratch.