
September 2023 Releases
The Regal Observability Dashboard is built to give your team visibility into AI agent performance, so you always know what your agents are actually doing once they're live.
Customer expectations for AI experiences have risen fast. Callers in healthcare, insurance, and financial services expect interactions that feel natural, stay on-script, and meet regulatory requirements. That means hallucination rates, guardrail breaches, and robotic language patterns are production concerns, and measuring them in real-time is what makes sustained improvement possible. With the Observability Dashboard, you can now track all these behaviors, across every AI agent, and every call.
.png)
The Regal Observability Dashboard is built around three things: monitoring what's happening now, making confident decisions about changes, and proving improvements over time.
.png)
Unlike platforms where surfacing insights requires manual sampling or custom reporting, every metric on Regal surfaces automatically across every call, in real-time.
When the dashboard surfaces an uptick in action errors, you can isolate exactly which actions are failing and drill into the transcripts to see the error detail. From there, you update the agent configuration, run tests before relaunch, and avoid the kind of prolonged impact that comes from incorrect action setup or timeouts going unnoticed.
For most teams, the latency view is about monitoring baseline performance over time. But when something shifts, the dashboard makes it easy to act fast. For example, when the dashboard alerts you to a latency spike on two specific agents, your team can drill into the turn-level transcript insights and see a consistent speech model delay across those conversations.
Rather than waiting on the vendor, you switch to a backup voice model while they resolve it. Customer impact is minimal, and you have a clear record of what happened and why. Looking ahead, that failover can be automated based on thresholds you define.
In rare cases, an agent may start narrating its internal actions rather than speaking naturally to the caller. It's not common, but when it happens it can persist for days without anyone catching it. The dashboard flags the robotic language rate for the affected agent so you can review the transcripts, update the guardrail prompt and test coverage, and close the loop quickly, before affecting your customer experience.
A business decision comes up: downgrade to a lower-cost model. You make the change and watch the dashboard. Over three days, the hallucination rate moves from 2% to 4%. That's evidence you can act on: your team can decide whether that's within acceptable range for the cost savings, or switch back to the more expensive model.
.png)
The dashboard tracks the full LLM quality, action adherence, and more across every call, in real-time.
The dashboard also tracks standard production metrics, including latency, sentiment, error rate, call time, and uptime, all surfaced automatically and linked to the individual call transcripts behind them.
Right now, it's never been easier to build an AI agent. With no-code builders, pre-trained models, off-the-shelf voice providers, you can have something running in a day. But it's also never been harder to make an AI agent enterprise-grade.
Most teams find that the build is the easy part. The hard part is everything after: guardrails, observability, knowing what your agent is doing at scale, and having the confidence to keep improving it without breaking what's already working.
Want to learn more about production-ready agents? Chat with our team.
An AI agent observability dashboard is a monitoring tool that tracks how an AI agent is performing in live production: including metrics like latency, hallucination rate, guardrail breach rate, and error rate. It gives operations teams real-time visibility into what their agents are doing and why, so they can identify issues and improve performance without relying on manual spot-checks.
Hallucination rate measures the percentage of calls in which an AI agent generated incorrect or fabricated information. Regal detects hallucinations automatically across all production calls, surfacing them in the Observability Dashboard without requiring manual sampling or transcript review.
A guardrail breach occurs when an AI agent violates a defined safety or compliance constraint — for example, straying from a required compliance statement, using prohibited language, or engaging with topics outside its designated scope. Guardrail breach rate is one of the key LLM quality metrics Regal tracks to help teams operating in regulated industries maintain compliance at scale.
With the Regal Observability Dashboard, every configuration change, prompt update, or model swap is reflected in live quality metrics. You can track how hallucination rate, guardrail breach rate, and other indicators shift in the days after a change, giving you a measurable record of improvement rather than anecdotal feedback.
In healthcare, insurance, and financial services, AI agents operate under strict compliance requirements. A hallucinated fact, an off-script statement, or a missed compliance disclosure can have real regulatory and customer consequences. Observability lets teams in these industries continuously verify that their agents are performing within required parameters, and show auditors and internal stakeholders the data to prove it.
Ready to see Regal in action?
Book a personalized demo.



