Training AI agents for real-world legal work. From reasoning to execution.
We provide long-horizon legal workflow datasets that agentic AI models fail at. Environments where agents have to use multiple tools to navigate complex contexts, documents, and make decisions to accomplish real-world legal objectives.
Backed by a network of hundreds of senior lawyers, with 9+ years of experience each in legal workflows that involve multinational agreements, litigation, privacy, compliance, intellectual property and 10 additional categories.
Infrastructure for AI legal hill-climb
The Revisioning platform streams and anonymizes daily real-time private data from its network. We filter to tasks frontier models have never been exposed to, with a 0% success rate across multiple passes. We deliver reproducible environments whose seeds average 10–20 legal reasoning steps and dozens of tool calls, each reviewed by domain experts against its rubrics. Every task preserves the expert solution, rubric, rationale, and reasoning traces, with deterministic and programmatic grades on every trajectory.
São Paulo • San Francisco