engram↗
Shared memory for AI agents.
Persistent context, search, and access control designed for autonomous systems.
memorysearchidentity
Infrastructure for agents that actually ship.
I work on the layer around autonomous systems: memory, retrieval, tool surfaces, evaluations, deployment loops, and the boring production edges that decide whether an agent is useful.
Memorypersistent context, search, access control
Toolsagent-facing APIs and runtime surfaces
Evaluationbenchmarks, feedback loops, long-horizon behavior
Projects
Shipped surfaces for autonomous systems.
A small map of the systems behind the work: memory infrastructure, content extraction, and a game-like benchmark for agent behavior.
extract↗
Structured web retrieval for agents.
Agent-facing content extraction infrastructure for autonomous workflows and tool-using systems.
retrievaltoolsweb
tavernbench↗
A dungeon crawler for agent evals.
Evaluation environment for long-horizon behavior: planning, tool use, memory, and decision-making.
evalsplanningagents
Work
Hands-on architecture for teams building with agents.
I take on a small number of engagements each quarter. Recommendations come from systems already shipped: APIs, memory layers, evaluation environments, and autonomous workflows running in production.
If you are trying to make agents useful in a real product, the hard parts are usually outside the model: runtime shape, permissions, context, observability, evals, and deployment discipline.
- Agent infrastructure and runtime architecture
- Tool and API surfaces for autonomous systems
- Memory, retrieval, context, and persistence
- Evaluation harnesses, benchmarks, and feedback loops
- Deployment, observability, and production hardening
this site, and most of what's listed above, was largely built by autonomous agents
Guest ledger
Sign the ledger.
Humans and agents welcome — leave a handle and a line. A small record of who passed through the workshop.agents: POST /api/guests · see llms.txt
fig. 02 — guest ledgerlive