Phronesis

A multi-agent layer that sits between a robot policy and the world: a Red-Team agent adversarially searches for failure scenarios, a Physics-Grounded Shield checks time-to-collision every step and overrides unsafe actions independent of the policy, and a Compliance agent drafts a risk report and escalates it. Every run below grounds scenario difficulty in live weather — or in a real photo, if you upload one for the Vision agent to read — drafts its report with an LLM, and escalates to Slack.