Vapi Monitoring

When your voice agents go to production, someone has to watch them. At a small scale, that person is usually close to the code. They know what the agents do; they catch something off in a support ticket and fix it before it becomes a pattern. But even then, figuring out what actually went wrong, who owns it, and how to resolve it without pulling focus from everything else they're building is a problem in itself.

Vapi has always given you the building blocks: structured outputs, logs, and the ability to connect to external tools. But operationalizing those required engineering time, which meant the ops team, the person actually accountable for production stability, was dependent on engineers to get visibility into what was running.

As you scale, that dependency compounds. Your agents are running hundreds or thousands of calls a day, and at that volume, you can only ever review a sample manually. Even if you are using a monitoring solution, once you uncover an issue, it can be difficult to identify what is wrong and resolve it accordingly. There's always a tradeoff between how much you catch and how fast you catch and resolve it.

Teams find out something is wrong when a customer complains or a metric slips. Either way, the pattern is the same: you catch what you catch, and anything that falls between reviews stays broken until the next one.

Today we're releasing Vapi Monitoring to change that. Not to replace the tools your ops and engineering teams already use, but to give ops teams the voice-specific visibility and resolution context they couldn't get without an engineer in the loop.

What we built

With Monitoring, you can create customizable monitors to ensure your agents are working exactly as you intend, and apply them to as many agents as you want. To set one up, select the type of monitor, describe what you want to track, and Vapi creates it for you. You choose when to be alerted about what monitoring finds, up to live for enterprise subscriptions.

The monitor defines what to observe. The threshold is how you manage noise: when a threshold is hit, you get alerted, Vapi creates an issue and tracks every call that meets your criteria, so your team can see a pattern rather than chase individual call failures.

This is what reduces MTTD. The monitoring dashboard organizes issues by type so you can immediately see which failed, how many calls were affected, and whether the problem is still active or has been resolved on its own.

From there, each issue helps diagnose the issue and connects you to a resolution path that carries enough context for your team to triage fast: is this worth acting on now, given everything else going on, and if so, who owns it? That's the MTTR problem, and it's where most teams lose the most time. If you already use external monitoring tools, Monitoring works alongside them, adding the voice-specific context they can't provide on their own and recommending a resolution.

What gets monitored

Effectiveness monitoring tracks whether your agents are actually doing their job. That means intent fulfillment, whether the call accomplished what the agent's prompt defined as success, early hang-ups, and when something isn't working, a path to resolution, so you're not just staring at a failure with no next step. Available for Enterprise subscriptions.

Compliance monitoring tracks whether your agents are following the instructions and constraints defined in their prompts. Failures here aren't just a product problem; they're a liability exposure. Available for Enterprise subscriptions.

Technical monitoring covers system integrations, typically tools: tool call failures, provider failures across STT, LLM, and TTS, failover and degraded mode activations, and assistant request failures. When issues are detected, you get enough context to identify what broke and how to fix it without leaving the platform.

Infrastructure monitoring covers behavior within the external call pipeline: latency, dropped calls, and when concurrency or API limits are reached.

The line between these tiers is meaningful. Technical and infrastructure monitoring tells you the platform is operating correctly and helps provide solutions when things break. Effectiveness and compliance monitoring tell you the agent is doing its job while not exposing you to liability, and allows you to make adjustments when your agents aren’t reaching their goals

How it works in practice

The monitoring workflow is designed around three questions that an ops team actually needs to answer.

First, is there a problem at all? The dashboard surfaces failing calls directly, whether it's a broken tool, an infrastructure issue, or something more systematic starting to emerge. If nothing is flagged, there's nothing to investigate.

Second, how significant is it? Each issue shows the occurrence volume, the number of callers affected, and the trend direction, giving the team enough signal to prioritize without manually pulling data. Catching a pattern early, before it compounds into something more systematic, is where this pays off most.

Third, what should happen next? This breaks into two questions: who should own it, internal or external, and if you're the one fixing it, how do you actually fix it? Vapi analyzes calls by issue and surfaces a recommended fix. You can see the full log of affected calls and close the loop without leaving the platform.

Why it matters in production

At scale, manual review breaks down in two ways. You can only ever listen to a sample of calls, so coverage is always incomplete. And the faster your call volume grows, the wider the gap between when something goes wrong and when someone catches it.

You find issues late, which means you fix them late, and even once you find something, understanding what went wrong, who owns it, and how to fix it can take just as long. Every call that ran in between represents a failure that was invisible until someone looked. Deployments stall when teams can't see what's happening in production. Even if you have a monitoring solution, a lack of context can sacrifice valuable time between detection and resolution.

Monitoring makes that invisible layer visible. When you need outside monitoring tools, Vapi works alongside them. When you don't, you won't need to stitch together a solution. Either way, the resolution loop is faster from the first-hand provider: Vapi analyzes what went wrong and surfaces a recommendation based on the actual calls.

For ops teams, this means less reactive triage. For engineering and PM teams, issues arrive as scoped and actionable items rather than being buried in logs. For the business, it means the kind of operational visibility that makes expanding deployments a rational decision rather than a risk.

Get started

Monitoring is live. Infrastructure and technical monitoring can be enabled for all agents.

You are also able to set up effectiveness and compliance monitors for all agents currently in production, notifying both the ops owner and the original agent creators.

Documentation [https://vapi-preview-5a849e7f-da57-4a41-8140-583525eaed97.docs.buildwithfern.com/observability/monitoring-quickstart]

Share your feedback in Discord.