[ Focus area ]
Agent planning
Graph memory, world models, and search — making long-horizon agents deterministic and auditable.
Research & open source
Reliable agents are a research problem. We work on it.
The gap between an impressive demo and a dependable system is where our research lives: planning, memory, world models, and evaluation for long-horizon agents.
GATS: Graph-Augmented Tree Search with Layered World Models for Efficient Agent Planning
In plain English: today's AI agents tend to wander — they re-think the world from scratch at every step and drift on long tasks. GATS gives an agent a map (graph memory), an imagination with levels of detail (layered world models), and a disciplined way to look ahead (tree search). The combination is deterministic: the same situation produces the same plan, which makes agent behavior auditable — the property production deployments actually need.
@article{williams2026gats,
title = {GATS: Graph-Augmented Tree Search with Layered World Models
for Efficient Agent Planning},
author = {Williams, Maureese},
journal = {arXiv preprint},
year = {2026}
}
Open source
Frameworks in the open
We build in public where we can. Open-source releases of our agent-planning and training tooling are being prepared — this page will grow.
[ Focus area ]
AI search & citation
How AI assistants choose what to cite — and the structured-data and content engineering that makes a business citable.
[ Focus area ]
Agent evaluation
Measuring whether an agent actually works: regression harnesses, calibrated contracts, and honest metrics.