Automated Replicate Failed Run Alerts
Every hour, WebRun opens Replicate, reviews recent predictions across your models, flags any that failed or errored out, posts the details to your engineering Slack channel, and mirrors anything that keeps repeating to Telegram so it's not missed.
How do I get alerted when a Replicate model run fails?
WebRun reviews your Replicate predictions every hour, flags any run that failed or errored, and posts the model, input, and error message to your engineering Slack channel. When the same error repeats more than twice in an hour it also pings Telegram, so a genuine pattern gets noticed fast instead of scrolling past in a busy channel.
- Failed runs get posted to Slack within the hour they happen
- Repeating errors get escalated to Telegram instead of getting buried
- Every model in the account gets checked, not just the main one
Built for ML engineering teams · AI product teams · platform engineers · indie developers
What does WebRun do on every run?
The exact actions WebRun takes, in order - in plain language, so you can adjust anything.
-
WebRun signs in and gets to work
Opens
replicate.comin a real browser with your saved login - no setup, no API keys. -
1
Replicate - check recent predictions for failures
replicate.com
WebRun checks recent predictions for failures in Replicate. - Open the Replicate dashboard and review recent predictions
- Filter to runs that failed or errored
- Capture the model, input, and error message for each
Done when Every failed run in the window is captured with its error.
-
2
Slack - post the failure details
slack.com
WebRun posts the failure details in Slack. - Post each failed run to the engineering channel
- Include the model name and error message
- Group repeats of the same error together
Done when The team has every failed run posted to Slack.
-
3
Telegram - mirror repeating failures
telegram.org
WebRun mirrors repeating failures to Telegram. - Mirror any failure that repeats more than twice in an hour to Telegram
- Keep the message short: model, count, and error
- Stay quiet when nothing crosses that repeat threshold
Done when Telegram only pings for failures that look like a pattern, not a one-off.
How is each run configured?
Secure by default
Connect once, stays signed in
WebRun signs in once and keeps each session in a persistent environment, so every run picks up right where it left off.
Every action is checked against this policy before it runs.
Questions, answered
Will it retry or restart a failed prediction?
No. WebRun only reports failures. Retrying or changing a model run stays a manual decision for your team.
What counts as an urgent failure worth a Telegram ping?
The same error repeating more than twice within an hour. That pattern usually points to a real problem, not a one-off bad input.
Does it check every model in my account?
Yes. It reviews predictions across every model you've run through Replicate, not just one.
Related templates
Put this on autopilot.
Turn it on in minutes - or have our team set it up for you.