I have been working on various versions of what I call my “second brain” since forever.
In the past, I typically used Markdown files inside a repo. This was a very easy way to store info and get to it from wherever I was. Management of the repo was pretty manual. You can use Obsidian for example to have a pleasant interface, vi, VS Code, whatever. Looking for a thing, “find all…”
It was a slight tweak on something I saw aa colleague do back in the 90’s, his system was similar, text files and grep. He use RCS for storing diffs of work (there was no git back then.)
Along come LLM models. Woot! That repo just became very cleaver.
The simplest extension that is very useful - you have all your data, point Agent chats to the files, I add AGENTS.md files in various directories to add context, this way a chat has a better idea of what you are talking about.
The AI experiments of late, basically I have this concept of an Inbox, you either put files there or paste data into a simple chat box. The agent has a quick look at the shape of the data and does some preprocessing. Info gets summarised and dropped into Markdown files with some YAML front-matter - this is so I can view using Obsidian.
I saw Google posted a nice version of this idea:
https://github.com/GoogleCloudPlatform/generative-ai/tree/main/gemini/agents/always-on-memory-agent
OpenClaw does a neat job with memories, the daily files etc and a consolidated MEMORY.md file. I like the accessibility of it all, if it is wrong just edit it.
I’ll clean up my latest edition and drop a link to the repo.
Links:
- Obsidian (point a vault to a folder) - https://obsidian.md/
- The always on agent example - https://github.com/GoogleCloudPlatform/generative-ai/tree/main/gemini/agents/always-on-memory-agent
- OpenClaw - https://openclaw.ai/