Memos as a shared long-term-memory solution for LLM's

[Feuerhamster]

Many AI providers have this "memories" system, where an LLM can save information snippets and recall them later or in other conversations.
I really like that system and the philosophy behind that, but one issue I had was that I use different providers (including self hosted stuff), different clients and never had any one of them access to the memories of another.

There are existing self hosted shared memory solutions, but I thought that my already running memos instance would be perfect for shared memories across systems. It has a great UI, supports markdown, and is the same app I use daily for note-taking anyway.

What my setup now looks like:
I made use of the new "spaces" feature and created a new space for LLM memories. Integrating with AI providers is plug-and-play due to the fully featured MCP. There is now a separate user account, just for the AI. This ensures safety and responsibility. I obviously dont want an LLM to read all my private notes, and also want to know which content was inserted by the AI and what is my own stuff. This way of access control is very capable, because I myself with my private account can just put a memo as shared or into the new space to make it available for the ai.
The entire thing got a system prompt with instructions on how to use tags to properly categorize topics and provenance.

Works very well, can recommend!

[aaronpowell]

I'm doing a similar thing - having an agent output to memos after doing some work so that we have a collaboration point to start with on future plans.