I kept running into the same friction: every Kimi session has an opaque UUID, I’d forget to write handoff files, and coming back to an old project meant reconstructing everything from scratch.
So I built a tiny tool that continuously checkpoints the meaningful parts of each conversation into a .kimi-memory.md file in your project root. It filters out tool noise, usage stats, and system prompts, and keeps assistant “think” blocks as [Realization] sections. If you’re in your home/Desktop directory, it writes to ~/.kimi/memory/general.md instead.
It wires into Kimi via hooks in config.toml (no core patches), so it just works after:
pip install kimi-memory
kimi-memory install
Then you can:
-
kimi-memory show: read current project memory -
kimi-memory list: see all sessions -
kimi-memory search “auth”: grep across every memory file
It’s idempotent, update-safe, and the memory files are plain Markdown so they’re easy to read, diff, or commit alongside your code.
If it sounds useful: github.com/Vibeypirate/KimiCodeMemorySystem
I’ve put this in the product feedback section cuz I’m hoping the Kimi team sees this and maybe works on an inbuilt memory feature that is actually better than this potentially lol, but for now its really useful for myself
I’m able to title sessions with /title “name” in the terminal and I can also do /sessions to see all of my previous sessions.