by
Artificial intelligence infrastructure startup Relai Inc. said today it has closed on $6.9 million in funding as it bids to ensure the reliability of autonomous AI agents for enterprises.
The company also announced the launch of its verifiable “continual learning” platform, which is designed to transform agents’s failures, traces, evaluations and also human feedback into reliable learning environments that can enhance their knowledge. The platform helps by identifying the root causes of AI agent’s mistakes and helps to rectify them by continuously optimizing prompts as well as their workflows, the tools they use, their contextual memory and more with live, in-loop regression controls.
The money came via two separate rounds, the startup said. Most recently, it raised $5.4 million in a pre-seed round of funding led by .406 Ventures with participation from the AI Tinkerers Fund and other strategic investors. Prior to that, it raised $1.5 million in what can only be described as a “pre-pre-seed round” led by Non sibi Ventures and Tedco.
Relai says the reliability of AI agents is more of a concern than ever before as enterprises look to move beyond the experimental stage and deploy them in production environments. But it remains one of the hardest unsolved problems in AI. Despite the best efforts of the world’s top AI model makers, the agents they power still suffer from unpredictable failures with alarming regularity.
Moreover, fixes often result in silent regressions that impact agent’s performance, leaving developer teams stuck in an endless cycle of patching, re-running evaluations and reactive debugging. The startup believes that one of the main causes of AI agent’s unreliability is that their learning is never verified against what already works. Its replayable learning environments aim to fix this by turning each failure into a reusable signal that can help to improve agents and then verify those upgrades.
Relai was founded by a leading AI researcher in Soheil Feizi, a Google Scholar and an associate professor of computer science at the University of Maryland who has contributed to more than 100 AI research papers. He graduated as a Ph.D. from the Massachusetts Institute of Technology and is also a recipient of the Presidential Early Career Award for Scientists and Engineers, which is the highest honor granted by the U.S. government to early-career scientists and engineers.
Unlike existing systems, which only check for regressions after modifying AI agents and shipping those changes into production, Relai aims to keep regression control within the optimization pipeline, Feizi said. Each proposed improvement will be continuously validated against a portfolio of prior environments as it’s being researched, rather than only being checked afterwards. Feizi calls this process “online, in-loop regression control,” and is key to enhancing AI agent’s capabilities without breaking them.
Another important difference is that Relai routes each fix to the correct layer of the agent’s stack. Depending on the nature of an agent’s failure, the fix might involve a prompt change, a tool wrapper, a memory update, a workflow adjustment, a code-level repair or a model-routing decision. The startup therefore diagnoses the root cause of each failure first, and then seeks to apply the smallest and most durable change at the appropriate layer.
Feizi said early adopters have shown remarkable improvements in agent performance. One financial services agent’s validation score rose from 39% to 80%, while a healthcare agent increased its performance from 62% to 96%. “For the past two years, the question was whether AI agents could use tools and pass benchmarks. They can,” Feizi said. “The real frontier now is whether agents can learn continuously from real experience without breaking what already worked. That is the gap Relai is closing… the missing outer loop that turns failures into durable, verified improvement.”
For convenience, Relai ensures its continual learning engine is compatible with leading agentic development frameworks via command line interface and workflow integrations. It can be used in combination with various AI coding agents, orchestration tools and enterprise AI stacks, enabling verifiable continual learning to be implemented with just two commands. It also provides a persistent system of record for each agent’s learning signals, optimization decisions and regression history, allowing developers to understand how its performance evolves over time.
Kevin Wang of .406 Ventures said shipping AI agents into production is no longer the biggest challenge for AI developers. “The hardest part is keeping it reliable as teams continuously improve it,” he explained. “Soheil has spent his career studying how AI systems fail and Relai turns that research into practical infrastructure that helps enterprise agents learn from failures without breaking what already works.”
Relai said its continual learning platform is open for limited-access onboarding now, with a broader public release set to come on June 22.
Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.
Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.
AWS’ powerful Graviton5 CPU makes its debut in new M9g and M9gd cloud instances
Lium raises $5.5M to unlock complex scientific data for AI models
Exclusive: MotherDuck adds agentic data ingestion to its cloud analytics service
Exclusive: Relai raises $6.9M to enable verifiable and continuous learning for AI agents
Zscaler unveils ZAgent Framework to automate zero-trust SASE operations
Jedify raises $24M to give enterprise AI agents the business context they lack
AWS’ powerful Graviton5 CPU makes its debut in new M9g and M9gd cloud instances
CLOUD – BY . 2 HOURS AGO
Lium raises $5.5M to unlock complex scientific data for AI models
BIG DATA – BY . 3 HOURS AGO
Exclusive: MotherDuck adds agentic data ingestion to its cloud analytics service
BIG DATA – BY . 4 HOURS AGO
Exclusive: Relai raises $6.9M to enable verifiable and continuous learning for AI agents
AI – BY . 4 HOURS AGO
Zscaler unveils ZAgent Framework to automate zero-trust SASE operations
SECURITY – BY . 5 HOURS AGO
Jedify raises $24M to give enterprise AI agents the business context they lack
AI – BY . 5 HOURS AGO
Forgot Password?
Like Free Content? Subscribe to follow.
Leave a Reply