Prefactor

Prefactor, made by Prefactor Pty Ltd and backed by Antler, Black Nova VC, and Func Ventures, is a runtime evaluation platform that continuously monitors and scores AI agent behavior in production, capturing every agent action as structured data and automatically enforcing policies or pausing risky actions. It offers runtime observability tracing every LLM call, tool invocation, and decision in real time, continuous evaluation scoring agents using LLM-as-judge, technical checks, and qualitative metrics, and enforcement that pauses or blocks high-risk actions with human-in-the-loop approval via SDK or API. Prefactor includes custom spans that enrich runs with context from external data sources like GitHub, databases, and APIs, agent lifecycle management that versions agents and gates promotion through dev, staging, and production based on evaluation performance, and native SDKs for TypeScript, Python, LangChain, Claude, Vercel AI, OpenClaw, and LiveKit. Pricing is freemium: a free tier includes 25,000 spans per month, with paid plans available for higher volume, and SOC 2 Type II and RBAC support in progress.