
Runtime Evaluation and Control for AI Agents
Check your Description
Prefactor is a runtime evaluation platform that reviews each step an AI agent takes in production and then holds, approves or blocks risky actions.
The platform records every run as traces and spans, scores runs with user defined evals such as LLM-as-judge and data risk checks, and sends held actions to humans for review.
The platform is built for AI platform teams moving agents into production, with TypeScript and Python SDKs and integrations for LangChain, Vercel AI and OpenTelemetry.
Prefactor was co-founded by CEO Matt Doughty and CTO Simon Russell and is venture backed.



