Key Takeaways
- Agent Registry gives leaders a living inventory to account for every agent application in Fiddler, with each agent’s model, tools, and deployment context.
- Each agent’s health, alerts, and token usage sit side by side, so teams can spot risky behavior and rising spend early and go straight to the root cause.
- The alerts you set in Fiddler roll up into the registry, powered by Centor evaluations that score jailbreaks, hallucinations, and sensitive data in real time.
- Agent Registry strengthens governance in the Fiddler AI Control Plane, connecting every agent to Fiddler’s audit trails, policy enforcement, and compliance evidence.
AI agents are quickly becoming part of how enterprises operate. In 2025, an estimated 25 million agents were already running in production, and by 2029 that number is projected to pass a billion. When agents deliver value, adoption moves fast, often faster than central AI and IT teams can track. As a result, most enterprises can’t say for certain how many agents are operating inside their organization today, what those agents are built on, or how they’re behaving.
That uncertainty falls hardest on leaders responsible for AI. Their goal is to give employees reliable AI systems and give agents the autonomy to drive the business forward, all within guardrails they can trust. Regulators, auditors, and the board increasingly expect them to know which agents are running, what they’re doing, and what they’re costing.
That’s why we built Agent Registry, a living inventory of your agents and the first thing you see when you log in to Fiddler. It gives leaders one centralized view for oversight, built to help them move from insight to action. Agent Registry bolsters governance in the Fiddler AI Control Plane, the system of trust that helps enterprises scale agents from development to production, and it connects every agent to the audit trails, policy enforcement, and compliance evidence Fiddler already provides.
Oversight and Accountability for Every Agent
Governance starts with seeing your agents in one place, and Agent Registry delivers that inventory. Every agent application and project in Fiddler appears in the registry, including coding agents like Claude Code through the Fiddler Claude Code plugin. As your AI footprint grows in every direction, the registry keeps it in one view, with each agent’s health, alerts, and token usage side by side, so leaders can stay ahead of two things that matter most as agents scale: how agents behave and what they cost.
- Agent behavior: Active alerts, including jailbreak attempts and sensitive data leakage, flag which agents need a closer look before small issues grow.
- Token spend: A sortable token usage trend for each agent shows which one is driving a spike without leaving the view.
From any agent, one click opens its full view in Fiddler, including the tools it uses and its context. Teams can review evaluator scores, trace individual decisions, and filter by segment to understand what changed and why. Evaluations and monitoring show what each agent did. Governance is what lets leaders act on it across every agent. Leaders get a portfolio-wide view built for oversight rather than per-project reporting, and the people closest to each agent get the detail they need to fix it.
Oversight Backed by Signals You Can Trust
Fiddler already lets teams set alerts on the evaluations that matter most to them, from jailbreak attempts and hallucinations to sensitive data leakage. Agent Registry brings those alerts together, giving leaders oversight of which agents are triggering them without checking each one separately.
Oversight is only as strong as the signals behind it. Those evaluations run on Centor, Fiddler’s evaluation engine. Centor Models run inside Fiddler and score every prompt, response, and tool call in under 80 milliseconds, with no external LLM judge adding a per-trace bill or seeing your data. For inputs outside what a specialized Centor Model handles best, Dynamic Model Escalation sends them to a Fiddler-hosted general-purpose model for a second look. In one reference dataset, that cut jailbreak false positives by 93.5%, so the alerts in the registry are ones leaders can act on.
This is just the beginning for Agent Registry. Seeing every agent in one place is the foundation, and it’s where we’ll keep expanding governance across the Fiddler AI Control Plane. The agents running in enterprises today are the fewest and simplest they will ever be. The teams that build governance now are the ones who get to keep scaling.
See what’s actually running across your enterprise. Request a walkthrough of Agent Registry.