Metis is an applied-research and product lab building the post-training and continual-learning layer that makes enterprise AI agents reliable in production. Its approach uses a company's first-party data, tools, and environments to shape agent capabilities and generate training signals through offline and online reinforcement learning.

Its first product, Insight, evaluates agents against real tasks to confirm they select the right tools, preserve context, and complete long multi-step workflows before deployment. The company positions itself as the missing learning and evaluation layer for dependable agentic computing, working with frontier labs and large enterprises.