Neel and Anay launched Sentrial, a production monitoring platform for AI agents that detects failures before users encounter them. The YC W26 startup automatically identifies loops, hallucinations, tool misuse, and user frustrations in real-time. When issues arise, Sentrial analyzes conversation patterns and model outputs to diagnose root causes and recommend specific fixes.

Building agents at SenseHQ and Accenture exposed the founders to the debugging nightmare many teams face. Without proper infrastructure, teams rely on logs and guesswork when agents fail, misclassify requests, or hallucinate content. A support agent might misclassify refund requests as product questions, or a drafting agent might produce confident but incorrect outputs from long specifications.

Sentrial works by wrapping clients with an SDK in just a couple lines of code. The platform detects drift including wrong tool invocations, misunderstood intents, hallucinations, and quality regressions over time. Teams see issues on the dashboard before customers file tickets. A quick MCP setup allows integration with Claude Code.

With a 14-day free tier and no credit card required, Sentrial addresses what the founders see as foundational infrastructure for serious AI products. The tool aims to eliminate the blind spots that plague production AI deployments.