Enterprise RL is a State Estimation Problem
Our technical direction for learning policies of dynamic enterprise workflows
Technical MemoComing soon
research
Automation requires agents that learn and specialize from experience. We study the foundations of these systems. Read our work on state estimation, world modeling, continual learning, and long-horizon decision-making.
JASPER: Learning More from Traces
A JEPA-inspired world model architecture built for digital workflow traces
ResearchComing soon
Atlas v0: Beyond Retrieval Benchmarking for Agent Memory
Toward Comprehensive Evaluation of Long-Horizon Agent Memory
Memory Is Not Retrieval
Retrieval answers questions. State determines behavior.
- Coming soon
SOLAR: Dynamic Residual Connections
Toward hyper connections as a pooled capacity market
- Coming soon
λBench: Measuring Memory via LLM-as-an-Interpreter
Can agents do repeatable tasks consistently?