data for frontier models deploying into open‑ended environments

Yes. We can already generate thousands of hours of data, and are scaling our infrastructure and tooling toward tens of thousands of hours per month. If your team needs a larger volume, please reach out.

Yes. We build custom data around your requirements, including new environments, traces, and trajectories across the applications, domains, and workflows you care about. We're already working with teams building benchmarks, evals, models, robotics, and manufacturing. Tell us what you need and we'll scope it with you.

Fig's datasets bring first in class learning signals that cannot exist in scripted tasks. Through deep observational telemetry across real work as it unfolds, the corpus captures real-world, open-ended workflows executed in continuously evolving environments. This provides customers with data that represents the exact type of environments and operations their models will be deployed to in production, which scripted collection can't reproduce.

Every trajectory is captured using Fig's proprietary capture tooling from consenting operators completing real work across research, operations, and engineering. There are no assigned tasks, prompts, or scripts, each recording contains the full state of the computer as the work happened, interruptions included.

Yes. Our data is built to be used by frontier teams across a wide variety of training applications. We're happy to walk through how this data or one of our custom solutions can fit into your existing pipeline. Tell us what you're building and we'll go from there.