See how every AI agent performs, and make it better
Prefactor evaluates, observes, and improves every agent in production, in real time, across every framework and workflow, with the runtime controls to act on what it finds.
TypeScript & Python SDKs · OpenTelemetry ingest · native for LangChain, Claude, Vercel AI, OpenClaw & LiveKit
Trusted by teams at
One loop, three stages, eleven capabilities.
Every capability on this page sits under one of three stages. Observe and evaluate build the evidence; act is the enforcement layer that uses it.
Observe
Every span, every agent, every run, as it happens, not reconstructed from logs after the fact.
Agent RegistryOne inventory for every agent: see what you have before you evaluate or enforce anything.Real-Time TracingEvery model call, tool call, and custom span, as it happens, not reconstructed afterward.Cost TrackingAttribute every token, API call, and compute cycle to the agent that spent it, in real time.Immutable Audit TrailEvery agent action, logged in real time and queryable. Your compliance evidence, always ready.Evaluate
Score outcomes, track drift from custom spans, and roll risk into 16 reviewable categories.
Quality AssessmentKnow when agent quality degrades, before your users do.Risk MonitoringRisk scored on two axes (data sensitivity and action consequence) reviewed in one place, not scattered across dashboards.Data TaggingTag anything in an agent's data, then track it, find it, or delete it everywhere it appears.Act
Turn what observe and evaluate find into a decision (block, delete, escalate, or stop) in milliseconds.
Runtime PoliciesAct on what evaluation finds: enforce rules at the agent execution layer, in milliseconds.PII DeletionTagged PII, gone everywhere it touched an agent, in one action.Human-in-the-Loop ControlApproval routing that pauses a high-risk action and puts the right human in the loop, with full context.Kill SwitchStop an agent immediately: one action, no code deploy.See it in action
One layer across every framework, from the agent registry to runtime enforcement. Watch a run get scored, flagged, and acted on.
Unified performance platform for agents, authentication, and risk management
Native SDK integrations for the frameworks you build on.
Connected through native SDKs, OpenTelemetry, and a TypeScript & Python core SDK that instruments anything else.
Not on the list? The TypeScript & Python core SDK and OpenTelemetry ingest cover anything else.
Your data lives where you need it to.
Prefactor's primary infrastructure runs in Australia. For enterprise engagements we deploy where your data needs to live: your region, or your environment.
Primary infrastructure runs in Australian data centres, which is where your data sits on a standard engagement.
Enterprise engagements can run on infrastructure stood up in your jurisdiction, so run data stays in-region.
Where the engagement needs it, deployment into your own environment is on the table. Talk to us about what yours needs →
Have visibility, so you understand what’s going on. But it’s got to be actionable — it’s got to deliver outcomes. Knowing what tools are doing what is one thing, but what can I do with that?Security Lead, Global Wagering Platform