Fiddler AI Introduces Agent Registry for Enterprise AI Governance
Fiddler AI has introduced Agent Registry, a centralized inventory designed to help enterprise leaders track every agent application in Fiddler, including its model, tools and deployment context. The registry brings agent health, alerts and token usage into one view within the Fiddler AI Control Plane.
- Agent Registry gives leaders a centralized inventory of agents in Fiddler.
- The registry displays agent health, alerts and token usage together.
- Centor evaluates prompts, responses and tool calls in under 80 milliseconds.
- Fiddler AI says Dynamic Model Escalation cut jailbreak false positives by 93.5% in one reference dataset.
- The registry connects agents to audit trails, policy enforcement and compliance evidence.
A portfolio view of enterprise agents
Agent Registry is intended to address a basic governance problem: enterprises may not know how many agents are operating internally, what they are built on or how they are behaving. Fiddler AI said an estimated 25 million agents were already running in production in 2025, with that number projected to pass a billion by 2029.
Every agent application and project in Fiddler appears in the registry. That includes coding agents such as Claude Code through the Fiddler Claude Code plugin. Leaders can see each agent's health, active alerts and token usage side by side, creating a portfolio-wide view rather than requiring them to inspect projects individually.
Fiddler AI said the registry is the first thing users see when they log in to Fiddler. From an agent's entry, a single click opens its full view, including the tools it uses and its context.
Alerts connect behavior with cost
The registry is designed to help teams identify both risky behavior and rising usage. Active alerts can highlight jailbreak attempts and sensitive data leakage, while a sortable token usage trend shows which agent is driving a spike in consumption.
Teams can open an agent's detailed view to review evaluator scores, trace individual decisions and filter by segment. According to Fiddler AI, this gives leaders a broad oversight view while providing the teams responsible for individual agents with the detail needed to investigate changes and fix problems.
Fiddler AI's existing alerts can cover evaluations including jailbreak attempts, hallucinations and sensitive data leakage. Agent Registry rolls those alerts up so leaders can see which agents are triggering them without checking each agent separately.
Centor supplies the evaluation signals
The alerts in Agent Registry are powered by Centor, Fiddler AI's evaluation engine. Centor Models run inside Fiddler and score every prompt, response and tool call in under 80 milliseconds, according to the company. Fiddler AI said the process does not use an external LLM judge that would add a per-trace bill or see customer data.
For inputs outside the areas best handled by a specialized Centor Model, Dynamic Model Escalation sends them to a Fiddler-hosted general-purpose model for a second look. In one reference dataset, Fiddler AI said this reduced jailbreak false positives by 93.5%.
Agent Registry connects each agent to Fiddler's audit trails, policy enforcement and compliance evidence. The company describes the registry as a foundation for expanding governance across the Fiddler AI Control Plane as enterprises scale agents from development to production.
