Training agents to combine knowledge in weights, text memory, and search on realistic legal work.
Scaling compute on your context.
