Deciduous.

It's a way of working/tools for working with an LLM that allow you to track decision tree graphs, have the robot make more informed decisions and build its own logical chain for history keeping, and modeling all the work as a DAG of events, goals, outcomes, decisions, and observations that network together to allow you to work better/smarter/faster, giving it a living and recorded memory and ways to explore all this.

It's easiest to check out the short demo on the site.

It also links to the live graph of how the tool has built itself.

http://notactuallytreyanastasio.github.io/deciduous/

Really interesting to see, because about 2 months ago I had a very similar idea, just with a bit more opinionated shape of the graph and context building, but more focused on the research and decision-making part. I started the development, but my focus eventually landed on a completely different aspect of that system.

So I have to ask, how well do you see it performing so far with regard to actually sticking to the data present in the system? Do you find the AI agents to adhere properly to the existing data?

On a similar note, can the "consensus" of the system be adjusted in a way where we keep the knowledge which was true at time T (decision provenance), but we avoid having that bit of information affect current decision making?

Cool name, with both hints at "decid[e]" and the graphs.

I'd be interested in integrating this with bug systems of decisions / goals, with actions being comments on those bugs (for work purposes) instead of having a custom deciduous-only DB.

Is this meant to be open source? I don't see a LICENSE.