Error monitoring where the AI agent fixes the bug and shows the proof.
One script tag tracks the errors your users hit and what they did. Server logs from Azure Event Hub and AWS CloudWatch are read for real errors, and uptime checks run every minute. Then the agent acts as your AI SRE: it finds the cause with AI root cause analysis, fixes the bug in a sandbox and opens a pull request with measured results.
How it works
From error tracking to a pull request with proof.
Capture
One script tag records console output, errors, failed requests, rage clicks and an annotated screenshot. Server logs are read every hour.
Triage
Errors are deduplicated by fingerprint and land in the Signals Inbox with a severity, next to slow pages and security findings.
Diagnose
AI names the root cause, the suspected files, a confidence level and whether it is code or configuration.
Fix
Create a ticket and the agent builds the fix in an isolated sandbox, replaying the user's steps in a headless browser to check it.
Prove
Your tests, typecheck and lint run before and after. The numbers go on a draft pull request, and review comments trigger follow-up commits.
What you get
Everything in Monitoring.
Error tracking from one script tag
A bug report button plus console output, JavaScript errors, failed requests, rage clicks, browser details and an annotated screenshot. Password fields are blurred.
Server log sources
Azure Event Hub and AWS CloudWatch, read every hour with your own filters. Errors and crashes become issues, deduplicated by fingerprint.
Signals Inbox
Errors, security findings and slow pages in one list with a severity. Create a ticket to have the agent work on one, or mute it.
Uptime monitoring
Every URL is called once a minute. Three failures in a row mark it down and open an issue. Critical checks follow your SLA escalation.
Performance
Page load, layout shift and interaction delay, with slow pages raised as signals.
Endpoint scanner
The endpoints your app calls are checked for security headers, CORS, authentication, HTTPS, rate limiting and error disclosure.
AI debugging and root cause analysis
Root cause, suspected files, a confidence level and whether it's code or config.
AI bug fixing
A fix built and tested in an isolated sandbox, then opened as a draft pull request.
Verify with tests
Installs, typechecks, lints and runs your tests before and after the change, and flags tests that were weakened.
Reproduce in browser
Replays the user's steps in a headless browser, signing in with a test login you set, with DOM and video captures.
PR follow up
Review comments on the pull request trigger new commits on the same branch, with a reply.
Auto mode
Optional. The agent opens draft pull requests itself for low-risk issues. Weakened tests, a critical finding or a change that adds a credential send the run back to a person.
In depth
A closer look.
Proof, not promises
Your own tests, before and after, on the pull request.
Before the agent changes anything it installs, typechecks, lints and runs your test suite, then does it all again after the fix. The before and after numbers go on the draft pull request. Tests that were weakened, by removing an assertion or skipping a case, are flagged rather than counted as passing.
- Install, typecheck, lint and tests, twice
- Weakened tests flagged
- A draft pull request with the numbers
- Review comments trigger follow-up commits on the same branch
Signals Inbox
Errors, slow pages and findings in one list, with a severity.
Everything Monitoring notices lands in one inbox: errors from the script and the server logs, deduplicated by fingerprint; pages with slow loads, layout shift or interaction delay; uptime failures; and what the endpoint scanner found. Each has a severity. Create a ticket to have the agent or a person work on it, or mute it.
- Errors deduplicated by fingerprint
- Slow pages from real user sessions
- Endpoint scanner: security headers, CORS, authentication, HTTPS, rate limiting and error disclosure
- Create a ticket or mute, straight from the list
Auto mode
Optional autonomy, with the limits written down.
With auto mode on, the agent opens draft pull requests itself for low-risk issues instead of waiting for a ticket. The limits are fixed: weakened tests, a critical security finding or a change that adds a credential send the run back to a person, protected paths are never edited, and merging stays with a triage manager.
- Off by default, per project
- Low-risk issues only
- Three conditions hand the run back to a person
- Protected paths are never changed
Connects to
Your browsers, your clouds, your repositories.
Who uses it
Fewer surprises, whichever side of the app you sit on.
Questions
Monitoring FAQ.
What does the script collect?
A bug report button for users, plus console output, JavaScript errors, failed requests, rage clicks, browser details and an annotated screenshot. Password fields are blurred.
Can AI fix bugs automatically?
Yes, with a person in charge. For each error, AI names the root cause, the suspected files and a confidence level. Create a ticket and the agent builds the fix in a sandbox, runs your tests before and after, and opens a draft pull request. With auto mode on, it opens draft pull requests itself for low-risk issues. A triage manager always merges.
Which server logs can it read?
Azure Event Hub and AWS CloudWatch, read every hour with filters you write. Errors and crashes become issues, deduplicated by fingerprint so one bug is one issue. A shipper for logs on virtual machines is coming.
How do uptime checks work?
Every URL is called once a minute. Three failures in a row mark it down and open an issue. Checks you mark critical follow your SLA escalation by email or SMS.
Does the agent change production?
No. Fixes are built and tested in an isolated sandbox and arrive as a draft pull request. Protected paths such as CI workflows, infrastructure, environment files and secrets are never edited unless you allow a path.
How does it reproduce a bug?
It replays the user's recorded steps in a headless browser, signing in with a test login you set, and keeps DOM and video captures from before and after the fix.
What does auto mode do?
It is optional, per project. The agent opens draft pull requests itself for low-risk issues. Weakened tests, a critical security finding or a change that adds a credential send the run back to a person. Merging is always a triage manager's decision.
Is it an AI SRE?
It does the part of an SRE's work that starts with an error: it watches errors, server logs and uptime, names the root cause and fixes the bug in a sandbox, with your tests as the proof. It never changes production or your infrastructure.
Which apps does it work with?
Any web app: the script tag works with any framework, on any host. Server errors come from Azure Event Hub and AWS CloudWatch logs, and fixes open as pull requests in GitHub, Bitbucket or Azure DevOps.
What does Monitoring cost?
There is no seat charge. Diagnoses and fixes count toward AI usage, on the platform quota or your own Anthropic, OpenAI or Gemini key, with a monthly budget per project.
Part of one workspace