Launch HN: Laminar (YC S24) – We trained a model to catch silent agent failures

Laminar, a company from YC S24, has released a model called flow-1, which is trained with reinforcement learning to detect errors in agent traces. The model is 23x cheaper than GPT-6-sol and matches its detection quality. Laminar has also developed a platform called Signals, which uses the flow-1 model to analyze agent traces and return findings as JSON. The platform can also group similar signal events into clusters, making it easier to identify and fix recurring problems. Laminar is open-source and self-hostable, and the company makes money from managed cloud services.