Respan AI × YC

Respan P-1 is a small language model built for detecting and handling sensitive data inside AI applications, released by the team behind Respan's observability and evaluations platform for agents. It targets engineering teams who already run PII redaction, guardrails, or compliance filters in front of their LLMs and have had to choose between fast regex based tools that miss context, larger general purpose models that are too slow or expensive to run inline, and third party APIs that require shipping user data off box. P-1 is pitched at teams shipping agents into production who need privacy controls on every trace, prompt, and tool call without paying the usual latency tax. The launch fits into Respan, formerly Keywords AI Inc.'s broader push around self-driving observability and evals for agents . The company originally started as Keywords AI, building infrastructure for dynamic LLM routing, then rebuilt around proactive observability and evaluations for AI agents that help teams trace, evaluate, and improve agent behavior from development to production . A dedicated privacy model slots naturally into that pipeline, giving customers a way to scrub or classify sensitive fields at the same layer where traces, evals, and gateway calls already flow through Respan. Respan was founded by co-founders Andy Li and Raymond Huang, both mechanical engineering majors from UIUC , and went through YC W24 . Over the past year, the company has scaled up to more than 100 startups and enterprise teams , and in March 2026, Respan raised USD 5 million in seed funding to expand its team and enhance infrastructure dependability while growing its proactive AI observability platform . P-1 is the first model Respan is shipping under its own name, extending the platform from monitoring what agents do into actively controlling what data they see.