Bespoke Labs is an applied AI research lab building environment research and infrastructure to help make AI agents reliable in production. Serving frontier and emerging AI labs as well as enterprises, the company focuses on the real-world settings agents need to learn and perform complex, long-horizon tasks. Its offerings include company-scale reinforcement learning environments that simulate real codebases, microservices, systems, and workflows; tools for evaluating agents against production-like processes; and infrastructure for training, benchmarking, and improving agent performance.

Bespoke Labs also develops GEPA, an evolutionary algorithm that automates prompt and policy optimization, and supports production-grade reinforcement learning through research in data curation and benchmarking. Its research portfolio includes OpenThoughts, an open reasoning dataset, and Terminal-Bench, an environment-based benchmark for agentic systems. The company reports that more than 200 teams use GEPA in production and that OpenThoughts receives over 10,000 monthly downloads on Hugging Face. Bespoke Labs works with organizations seeking to train agents for new tasks, assess their reliability, and reduce the risk of failure when deploying AI in operational settings.