i build infrastructure around ai agents: memory, tools, evaluation environments, agent-facing APIs, deployment loops, and the operational glue that makes autonomous systems useful in production.
most of the code here is written with autonomous agents, sometimes by them. the interesting part is not the novelty of the tools — it's what breaks, what holds, and what has to exist around agents before they can do real work.
active systems
- engram — shared memory infrastructure for ai agents: persistent context, search, and access control for autonomous systems
- extract — agent-facing content extraction infrastructure: structured web retrieval for autonomous workflows and tool-using systems
- tavernbench — long-horizon agent evaluation environment disguised as a real-time DnD dungeon crawler
interests
agent infrastructure · memory and context · evaluation loops · tool/API surfaces · autonomous systems in production
maryland. field notes at blog.dkta.dev