Back to NewsroomBack to Newsroom
Netra launches Alerts to flag AI agent issues in real time
Share this

Share this news via

Netra launches Alerts to flag AI agent issues in real time

Netra, the AI observability, evaluation, and simulation platform by KeyValue, has launched Alerts. The new feature notifies teams when an AI agent crosses a set limit on cost, latency, error rate or token usage, so they can act on issues as soon as they occur.

Monitoring across four metrics

Alerts lets teams set rules on four metrics: cost, latency, error rate and token count. Each rule can track a full request (a trace) or a single step within it, such as an LLM call or a tool execution (a span).

Rules can be narrowed by model, tenant, environment or service. An optional time window helps prevent a single failure from triggering an alert. 

Notifications in real time

Netra checks rules against incoming traces as they arrive, with no polling delay. When a condition is met, a notification goes out to Slack or email, usually within seconds.

Each notification includes the value that triggered it, a timestamp and a direct link to the trace behind it. Teams can route alerts to the people responsible for them, sending cost alerts to one channel and error alerts to another.

AI agent reliability that doesn’t stop at launch

Having built and deployed more than 25 AI agents, KeyValue has seen firsthand how agent behavior can change when exposed to real users and real-world conditions. Building an agent that works is only part of the challenge. Keeping it reliable in production requires continuous monitoring and improvement.

That experience shaped Netra, a platform built to help teams maintain AI agent reliability throughout the agent lifecycle. Netra Alerts takes this a step further, by notifying teams when predefined thresholds for cost, latency, errors, or token usage are exceeded, helping them identify and respond to issues in production.

Learn more about Netra at getnetra.ai.