I built Khaos Brain because most AI memory features feel too shallow for real agent work.
Saving "remember this next time" is useful, but the more valuable unit is accumulated experience: what condition appeared, what action was taken, what result happened, which route failed, and which route later became reliable.
The problem is not that an agent cannot remember a sentence. The problem is that after doing similar work many times, its working experience often does not accumulate into something inspectable and reusable.
Khaos Brain is an open-source, local-first experience organization tool for AI agents. It stores experience as visible file-based cards instead of opaque memory.
What it does
The current release is Codex-first, but the idea is broader:
- before a task, the agent can retrieve relevant experience
- after a task, it can write back observations and lessons
- maintenance workflows can organize those cards over time
The cards can be searched, inspected, reviewed, diffed, merged, and rolled back with Git.
One part I care about is keeping personal memory and shared knowledge separate. Personal preferences stay local. Reusable task models, engineering lessons, and reviewed skills can be shared through a GitHub-backed organization knowledge base.
The easiest way to try it is to hand the GitHub URL to your coding agent and ask it to install and enable the project.
Repo:
https://github.com/liuyingxuvka/Khaos-Brain
Feedback:
https://github.com/liuyingxuvka/Khaos-Brain/discussions/2
I would especially like feedback on:
- whether "local-first experience system" works well as a framing
- whether visible experience cards feel more useful than opaque AI memory
- what kind of agent work would benefit most from accumulated experience
Top comments (0)