Relai Raises $6.9M for AI Agent Reliability

AI infrastructure startup Relai has closed $6.9 million in funding across two rounds, led by .406 Ventures, to address one of AI's most persistent problems: unreliable autonomous agents in enterprise production environments. The company simultaneously launched a verifiable continual learning platform designed to turn agent failures, traces and human feedback into reliable improvement signals. Founded by MIT-trained University of Maryland professor Soheil Feizi, Relai's platform diagnoses root causes of agent failures and applies targeted fixes across prompts, memory, tools and workflows using in-loop regression controls. Early results show dramatic gains, with one financial services agent improving from 39% to 80% validation scores and a healthcare agent jumping from 62% to 96%.