alfred(again)
A domain-agnostic multi-agent orchestration framework
Nerd out here →Problem
I love to cook. I hate planning what to cook. What if I could build an AI that handles meal planning, recipe management, and grocery coordination for me?
Built
The first Alfred didn't survive its own complexity — I learned more from killing it than building it. The second attempt became a five-agent orchestration system with deterministic state management, a three-layer context model, and a pluggable domain protocol — now being extracted into a standalone Python package.
Learned
Why deterministic systems matter when LLMs are nondeterministic. How to design abstraction boundaries that actually hold. That the hardest problem in orchestration isn't the agents — it's the state.



