Agents should learn from how work actually happened.
Every day, businesses close. When they do, years of operational history are usually deleted with them: every order, invoice, ticket and decision.
That record is exactly what AI agents need to learn how work gets done. Not a tidy simulation of a company, but the real one, with its backlogs, disputes, workarounds and people covering for each other.
Cogac acquires that record from businesses at the end of their life, protects the people in it, and turns it into environments that labs use to train and evaluate agents.
The name says what we build. Cogac stands for Cognitive Agent Contexts: the real working context an agent needs to learn how a job is done.
Principles
Real over synthetic
Provenance is the product. If an environment did not come from a real company, we do not ship it as one.
Privacy before fidelity
We never trade de-identification for realism. Structure and timing make data useful; identities do not.
Fair to the companies we buy from
Acquisitions return value to creditors of failed businesses, through formal processes and at fair prices.
Built with researchers
Interfaces, graders and splits are designed with the labs that use them, and published openly to customers.
We’re hiring.
Research, data engineering and acquisitions.
Give your agents real experience.
Evaluate environments for your workflows. Discuss coverage, delivery and commercial licensing with our team.
Get Access