The Pulse
Replit Lets Monitoring Agent Deploy Production Fixes
Replit’s Monitoring Agent can investigate production alerts, attempt and validate code fixes, and publish them automatically if users enable a separate experimental setting. The feature also lets users ask the agent about production logs an
AI.info Team · Written with AI · fact-checked ·
Three signals—application error logs, HTTP errors and request latency—can trigger an investigation by Replit’s Monitoring Agent. Announced on October 22, the tool can examine an alert, attempt and validate a code fix, and publish that fix automatically if the app owner enables the separate republishing control.
Replit authors Akram Hussein and Carl Cortright describe the product this way:
“Replit Monitoring Agent gives every internal app builder and entrepreneur an autonomous on-call engineer for their app.”
The feature extends Replit’s app-monitoring tools from alerting users about problems to attempting a repair. With alerts and Auto-fix enabled, the agent investigates in the background using production metrics, runtime logs and the app’s code. If it identifies an actionable cause, it tries a fix and validates the change. Replit cautions that an alert does not guarantee a fix; an issue with an external service, for example, may need attention outside the app.
More production information on request
The three signals monitored automatically are narrower than the information the agent can inspect on request. Users can ask about production logs and metrics without waiting for an alert or turning on Auto-fix. Replit’s documentation gives examples including investigating recent server errors, comparing endpoint latency, checking CPU and memory use, and reading logs for a deployment revision that failed to start.
Available telemetry depends on the deployment type and cloud provider. The documentation lists request counts, latency, availability and HTTP response codes; endpoint-level performance, including average and percentile latency; CPU and memory utilization, active instance counts, startup latency and instance uptime; disk usage and network egress; and application error counts and runtime logs. Owners can provide an approximate time, time zone and affected endpoint to help narrow an investigation.
Publishing remains a separate decision
Replit separates monitoring alerts, automatic fixes and automatic publishing into distinct settings. A user can receive alerts without allowing the agent to change code, or allow it to attempt repairs while retaining the option to review and publish changes manually. With Auto-republish enabled, a validated fix can go live automatically; otherwise, the user can review the result in Agent chat and republish the app themselves.
Auto-republish is experimental and requires Auto-fix. Replit says enabling Auto-fix alone does not authorize the agent to publish changes to a live app. The company also warns that some changes, including migrations and index operations, may be risky.
Owners configure monitoring separately for each published app. For an existing app, monitoring alerts, Auto-fix and Auto-republish can be adjusted in its publishing settings; users can leave alerts on while turning off automatic fixes, or keep Auto-fix on and disable automatic republishing. Monitoring Agent runs may incur agent-usage charges, which users can track on Replit’s usage page.
The Monitoring tool lists issues that are active and those that stopped affecting the app within the last 30 days. Users can expand an alert to see why it fired and how many monitoring passes detected it; alerts Replit cannot evaluate are marked “Cannot measure.” Email notifications have separate controls, including severity thresholds, and turning off emails does not turn off monitoring or Auto-fix.
From alerts to attempted repairs
Replit launched App Monitoring on April 29, introducing health checks and alerts for published apps. Its announcement also said Replit Agent could inspect production logs and read the production database with read-only access to help diagnose failures. The new Monitoring Agent announcement describes a broader response flow: investigate an alert, attempt and validate a fix, then deploy it if the owner has enabled automatic publishing.