Fix
An AI coding agent that fixes what your users hit.
When Sentry reports an issue or a customer emails about a bug, an agent wakes straight away. It checks the report is genuine, fixes the code, deploys, confirms the fix in production and resolves the issue.
An incident, fixed and verified
A Sentry issue that woke an agent: the legitimacy check, the fix, the deploy and the production check.
What an AI coding agent does here
An AI coding agent reads a codebase, makes changes and runs them. Most are pointed at tickets a developer writes. This one is pointed at production: the errors your users actually hit.
The hard part is not writing the fix. It is knowing whether to make it, proving it worked, and closing the loop with whoever reported it.
How the agents do it. Step by step, on a schedule.
A report arrives
A Sentry issue arrives by webhook, or a customer’s bug email is routed from your inbox. The incident job starts at once, not on the next schedule.
Is it real, and should we?
Before touching the repository the agent decides: not real, not in good faith, conflicts with a decision already made, or legitimate. A bad-faith request is refused and escalated.
Fix and test
It reproduces the problem, changes the code and runs your tests.
Deploy and verify
It ships the fix and checks production. If the fix is merged but not live yet, the issue stays open with a comment.
Close the loop
The issue is resolved only when the fix is verified live. If it cannot fix it, you get one task with what it found.
What you get. Small things that add up.
Wakes on the report
Webhooks and inbound email trigger the job immediately.
A legitimacy gate
“Please remove your licence check” is refused. “Your licence check rejects my key” is investigated.
Verified in production
A fix is not done until the live site proves it.
Judged separately
A separate verifier reads the record of what the agent actually did, not its own summary.
What it will not do. On purpose.
- It will not mark an issue resolved because a pull request merged. Only a verified live fix counts.
- It will not partially implement a request it judged to be in bad faith.
- It needs access to your repository and a way to deploy. Without them it reports what it found and stops.
Questions
Does it replace Sentry?
No. Sentry finds and groups the errors. OpenHelm acts on them. The two work together.
Can I review fixes before they ship?
Yes. Have it open a pull request and wait for your approval instead of deploying.
What about customer bug reports by email?
The support agent routes them to the incident job with the customer’s exact error, and the customer is told it is fixed only once it really is.
Which models does it use?
Your choice: your own Claude Code or Codex subscription, a local model, or hosted models. See Models.
Give your product a team that never clocks off.
Start free in the browser, or run it on your Mac with your own Claude Code or Codex plan.