We built peat, step 1 toward a better long-term memory system for serious, large, multi-agent projects (500k LoC+)
github.com/flowerornament/pe…
peat gives you:
- Quick CLI access to the full history of agent operations in your repo
- Agent briefs that help bridge context between many compactions
- Agent-authored memories ("observations")
- Log replay (revisit any past state)
- Vector search across agent sessions
- Time-frame zoom (day, week, month, quarter, year)
It's a simple design, heavily influenced by
github.com/VictorTaelin/OptM…, leveraging the affordances of
github.com/flowercomputers/b…
Notably, this was extremely easy to build. I've built simple versions of append logs with replay in Rust for other dev tools—bog makes it dead simple, performant, and fixes the tomb-stoning problem.
For me, a missing piece in my agent toolkit to go along with:
-
github.com/flowerornament/an…
-
github.com/gastownhall/beads
-
github.com/jj-vcs/jj
--
In the future, your whole digital life will be a log with materialized views.
Thanks
@flowercomputers