Laminar

Laminar is an open-source observability platform for AI agents, and this launch introduces flow-1, a model built specifically to read through agent traces and pinpoint where runs went wrong. It is aimed at teams shipping agents into production workflows where a single run can contain hundreds of model calls and tool invocations, making manual review impractical. flow-1 is a new model for understanding agent traces, trained with reinforcement learning to identify hard-to-spot failures and explain why they happened. The pitch is cost. On a benchmark of 523 difficult agent traces across coding, legal, customer support, CRM and more, flow-1 surpasses GPT-6-sol with high reasoning in failure detection quality, with Sol flagging more traces and making it noisier while flow-1 digs into the evidence to surface the failures that matter. At $0.05 per million input tokens and $0.30 per million output, flow-1 can process 23x as many traces per dollar as GPT-6-sol at comparable quality, and runs about 25% cheaper than GPT-6-luna with better detection and explanations. It runs inside Laminar's Signals agent, which can grep across spans, inspect them in full, and return structured JSON matching a user-defined schema, so findings drop straight into existing pipelines. The launch matters because sampling has been the default compromise in agent observability, and flow-1 is priced to let teams analyze every production trace continuously rather than a slice. Laminar was founded in 2024 by Robert Kim and Din Mailibay , went through Y Combinator, and raised a $3M seed led by Atlantic with participation from Y Combinator, AAL.vc, and angels including Ben Sigelman, co-creator of OpenTelemetry, and Ant Wilson, CTO of Supabase . Robert, who posted the launch, is co-founder and CEO.