Hyperbench builds AI agents for hardware-in-the-loop testing. The agents understand a team's system, work across repositories and verify their changes on real hardware. Users connect their code, configurations and test benches, and the agents carry an issue through to a verified change. The company is in a private beta with early access by sign-up.
The workflow has four stages. Understand: trace issues across code and hardware. Reproduce: capture the failure on the bench. Repair: delegate fixes across repositories. Verify: retest on hardware and review the draft PRs. Code, test results and recordings are reviewed together, so each handoff comes with the supporting results.
Hyperbench runs on a team's own infrastructure. Execution is local with shared context, so a team can share the system, and users bring their own bench by using the hardware they already have. The company describes a custom stack that stays under the customer's control.
The founders are Justin Leong, previously a software engineer at Anduril on Roadrunner GNC firmware, and Ivan Ge, previously a Member of Technical Staff at OpenAI working on software infrastructure. The company describes the team as working at the intersection of autonomous systems and AI infrastructure. Visitors can talk to the team or join the beta.