Show HN: IHMT – Infinite Hierarchical Memory Tree for AI Agents

That sounds interesting! Are there any plans to conduct quantitative comparisons with other agent memory approaches?

Hi

IHMT focuses primarily on simplicity, privacy, and the complete absence of external infrastructure.

I’d love to be able to make comparisons with other memory systems. Based on what I’ve tested, I can assure you that it saves a huge number of tokens. It’s like having the $200 plan but only paying $20. I haven’t spent any session tokens since I started using it.

I’d like to be able to compare it with other systems and publish the results.

IHMT is like infinite memory, because when you work on a huge project, the next time you open a new agent, it doesn’t need to read the entire project—just the memory, which is only a few tokens—and that’s enough for it to understand the project. The amazing thing about this is that it remembers everything, no matter how long ago the conversation, project, etc., took place. You can save the conversation and transfer it to another computer, and it only takes up a few KB.

Honestly, I don’t know how I was able to work before this system. Now I don’t have to search for the project I saved in Claude Code and pick up that conversation—I just open Claude Code, OpenCode, or whatever, and ask, “Do you remember when we were doing such-and-such?” And it simply remembers everything, no matter which directory you’re in. Best of all, since it’s a new conversation, it saves millions of context tokens.

Best regards