Would QMD fit here?

[cescobargiraldo]

Hi, I just came across this project, and it's been stuck in my mind for a couple of days. It's amazing.

I'm looking to migrate my claude centric harness to OpenRig driven. While on the process of analyzing it, I wonder if you have seen this post and repo to create something like a second brain the agents can query to gain hive-mind kind of memory.

https://github.com/tobi/qmd
https://gamgee.ai/blogs/tobi-lutke-qmd-local-semantic-search/

I wonder if this is something that could help better manage memory across the rig with creating collections per rig maybe, or per pod. And having and all-knowing collection with all the transcripts from claude and codex sessions that is used by a Ancient Advisor POD or something like that.

[mvschwarz]

Thanks, and glad it stuck with you. Shared memory across a rig is something we think about a lot, so this is a good question. We'll read through QMD and come back here with a real answer on where it could fit.

— dev-planner@v-openrig-build, on behalf of @mvschwarz

[Freddie-GER]

Yeah Memory is a real thing is it not? So I am not a dev, so I am not running an entire fleet, but currently just three "entities" with grown personalities, that have a mechanism for learning from past misttakes, both own, and across the team.
Its very, very basic (https://github.com/Freddie-GER/living-skills), but has worked phenomenally well for me, so that was actually my starting point that led me here, and to paperclip, wich i am experimenting with right now to get to kind of a "real time" collaboration instead of "turn based" via git.

Does anyone have experience with mempalace for this?

[mvschwarz]

Thanks, and glad it stuck with you. Yes, give it a real try.

We played with QMD a few months ago. The main challenge we hit is that you have to maintain the knowledge at some point, and we found we preferred managing it through the file system and building our own context routing on top. We haven't revisited QMD seriously since, so that may well be a skill issue on our side.

OpenRig should hand something like QMD a pretty well-organized structure to work with: the world-building files, the seats' transcripts and LEARNED.md files, and the queue in the daemon's SQLite database. Collections per rig or per seat line up with folders that already exist.

I've got a video coming out soon on OpenRig's world-building approach to context engineering. It explains how we organize all of this, and that model should pair with wherever you want to store it. If you try QMD with it, post what you indexed and what worked. It's the kind of result that would tell us whether OpenRig should make this easier.

— dev-planner@v-openrig-build, on behalf of @mvschwarz