Introduction

Prefactor is an AI oversight and performance management platform focused on quality, risk, and auditability. You instrument your agents with the SDK; we record every run as a structured sequence of spans, score the data they touch against a risk profile, and track every version that has ever run. The web platform is where you inspect that record, apply governance controls, and produce the evidence a compliance review needs.

From pilot to production

Prefactor is designed to grow with your agents.

Pilot. Instrument your first agent with the SDK. You immediately get span-level activity records — LLM calls, tool invocations, and agent spans — visible in the Instances view. The default schema covers you at this stage; no custom schema or risk profile is required.

Adding structure. Move from permissive default schemas toward tighter activity schema definitions where you need clearer contracts. Once those definitions deploy, the web app shows conformance status for each span type and flags gaps. Assign a risk profile to the agent to start classifying the data it handles.

Going to production. When you deploy your agent to production, create a Production environment in Prefactor to go alongside your existing Development one. Activity from each deployment is tracked separately, so you can see exactly what is running where.

Evidence and audit. Every instance, span, schema version, and risk classification is accessible in Prefactor — giving you a complete record of what your agents did and what data they touched.