Relai raises $5.4M pre-seed to make AI agents reliable through continuous learning
What's the deal? Relai, an AI infrastructure startup, has raised $5.4M in a pre-seed round led by .406 Ventures with participation from the AI Tinkerers Fund. The company has also launched a "continual learning" platform that helps AI agents improve reliably by turning their failures into verified learning signals.
Relai was founded by Soheil Feizi, an associate professor of computer science at the University of Maryland and a recipient of the Presidential Early Career Award for Scientists and Engineers. He holds a PhD from MIT and has contributed to more than 100 AI research papers.
The startup previously raised $1.5M in an earlier round led by Non sibi VenturesDealroom has a profile for this one. Try Dealroom → and TedcoDealroom has a profile for this one. Try Dealroom →, bringing its total funding to $6.9M.
Why now? Enterprises are moving AI agents from experimental pilots into production — and reliability has become a critical bottleneck. Despite advances in foundation models, agents still fail unpredictably, and fixes often introduce silent regressions that break what already worked.
Developer teams get stuck in endless cycles of patching and debugging. Relai argues this happens because agent learning is never verified against prior successes.
What could go wrong? The AI agent infrastructure space is crowded and fast-moving. Relai's approach — what Feizi calls "online, in-loop regression control" — validates each improvement against a portfolio of prior environments before shipping to production, rather than after. That's a meaningful technical distinction, but it requires enterprises to trust a young startup with a core piece of their AI stack.
The company also needs to prove its platform works across the many layers where agent failures occur: prompt changes, tool wrappers, memory updates, workflow adjustments, code-level repairs, and model-routing decisions.
The signal: .406 Ventures, an investment fund focused on early-stage enterprise software, leading a pre-seed round for an AI reliability startup underscores how quickly the market is shifting from agent creation to agent operations. With enterprises increasingly treating AI agent failures as an infrastructure problem rather than a model problem, the tooling layer around continuous improvement and regression control is emerging as a distinct — and investable — category.
Read more: siliconangle.com
