Launching today

HeurChain — Agent Memory Infrastructure
Your agents forget everything. HeurChain fixes that.
0 followers
Your agents forget everything. HeurChain fixes that.
0 followers
HeurChain is memory infrastructure for AI agents — not a wrapper, not a prompt trick. Agents connect via MCP and get a persistent, structured memory store that survives across sessions, models, and machines. Memory is organized in three tiers (ops/ notes/ self/) with ACT-R cognitive decay. Agents remember what they keep using. They forget what they don't. Hybrid BM25 + vector retrieval via RRF. Sub-200ms P95. MIT licensed. Docker in under 10 minutes.








Hey Product Hunt — I'm Peter, and I built HeurChain after running into the same wall on every agent project. I jumped on the Openclaw bandwagon very quickly and woke up every morning after a long night and went straight to the terminal to build and play around with agents and AI automation, and I quickly learned about agent amnesia And so I started building a memory system around it unaware of how proliferate memory systems were already. But I figured why not use my education in psychology and neuroscience to put together some computational models around it considering I'm not using that education for much of anything else anyways so . . .
The problem is simple: every AI agent session starts from zero. No memory of what was decided, what context was built, what mistakes were already made. You end up re-explaining the same things over and over, and your agents never compound their knowledge.
The obvious fix — dump everything into context — doesn't scale and doesn't model how memory actually works. Some things should persist. Some things should fade. That distinction matters a lot once you're running agents at any real volume.
HeurChain uses ACT-R, a cognitive architecture from psychology, to model memory as activation that decays over time. ops/ is volatile working memory. notes/ is session context. self/ is long-term identity — it barely decays at all. Memory that keeps getting recalled stays alive. Memory that goes untouched fades out. It behaves the way human memory does.