Training agents to work reliably for your customers.
About
Metis is an applied-research and product lab building proprietary intelligence: the post-training and continual-learning layer for enterprise agents.
Today’s information stack was built for static software and generic models, not autonomous systems operating real tools and workflows. The result is brittle agents, endless pilots, and low trust in automation. Metis supplies the missing learning and evaluation layer so agentic computing becomes dependable at scale.
What We're Building
Our first product, Mantis, turns a company’s first-party data, tools, and environments into a continuously learning agent stack:
Learns from your reality
Uses your first-party data, tools, and environments to shape agent capabilities.
Improves through practice
Generates training signals and runs offline/online RL so agents get better over time.
Validates before production
Evaluates agents against real tasks so they pick the right tool, preserve context, and complete long, multi-step workflows with higher reliability and throughput.
We focus on pragmatic progress: a data flywheel where agents learn from every interaction while remaining observable, auditable, and safe. We’re fortunate to be trusted by frontier labs and Fortune 500 enterprises.
Interested?
Request access to our private beta below.