Bumblebiber

4 ms·

1. 1.▲
   by Bumblebiber 7mo ago

   Thanks <3 I designed it to be very token saving. It loads mostly the very brief summaries of each memory (and there's an algorithm which filters out the lesser important ones). Only if the agent need to know more details, it will di
2. 2.▲
   Show HN: My OpenClaw knows EXACTLY what it did a week ago. Thanks to "hmem"-MCP

   1 points by Bumblebiber 7mo ago|2 comments
3. 3.▲
   by Bumblebiber 7mo ago

   Some core features worth calling out: 5-level lazy loading instead of flat RAG. On spawn, agents receive only L1 — one-line summaries of every memory entry. When something looks relevant, they drill into L2, L3, etc. on demand. This
4. 4.▲
   Show HN: Hmem v2 – Persistent hierarchical memory for AI agents (MCP)(github.com)

   2 points by Bumblebiber 7mo ago|4 comments
5. 5.▲
   by Bumblebiber 7mo ago

   Author here, happy to answer questions. Some background: I run a multi-agent AI system (orchestrator + specialized agents) across multiple machines. Two things kept biting me: 1. *Context dilution:* In long sessions, earlier context gets co
6. 6.▲
   Show HN: Hmem – Persistent hierarchical memory for AI coding agents (MCP)

   2 points by Bumblebiber 7mo ago|3 comments