Investigate incidents while your team is offline

How Novix gets involved

A customer report or an auto-summon.

A webhook starts a run the moment it lands. Sources without one are polled every five minutes by default, so the gap between a report arriving and a run starting is minutes rather than a working day.

How the agent handles it

One run per task, across every worker
A run claims the task with a single conditional write, and a second worker handed the same task returns having spent nothing. A live run re-stamps its claim, so a long night never costs you a second diagnosis.
Your own build and your own tests
BuildClones your repo into an isolated container, applies the patch, and runs your own build and tests, so a fix that does not compile says so before you approve. For a change you can see, it also starts the app in that container and photographs the page before and after. Always runs.
The patch is applied in a throwaway container and the repository build and test commands run against it. There is no per-team off switch, because a check a workspace can quietly disable is not a check the product can promise.
One repair attempt on a build that failed
A patch that actually failed to build is re-drafted with the failing log in hand, on its own budget, then re-verified. A repair that does not work changes nothing: the original patch, verdict and trace stand.
The ceiling nothing turns off
A pull request is the most that happens unattended. Five things decide whether it lands ready for review or as a draft carrying the reason it was downgraded, and no path in the codebase merges.

What your team gets

A diagnosis waiting when somebody wakes up
A pull request already through your build and tests, or a draft saying why it is one
The build log, and an honest could-not-run when the toolchain is not in the image

Novix never merges, and there is no setting for it

Merging is shipping, because your main branch deploys. A person approves, and on the dashboard that takes two clicks and the capability to approve fixes.

Works with

Slack
Discord
Forwarded email
Sentry

Try it on a real customer problem.

Connect the place reports or product signals already arrive. The same agent takes it from evidence to a tested pull request.

Start free Chat with the founder

How the agent handles it

One run per task, across every worker

A run claims the task with a single conditional write, and a second worker handed the same task returns having spent nothing. A live run re-stamps its claim, so a long night never costs you a second diagnosis.

Your own build and your own tests

The patch is applied in a throwaway container and the repository build and test commands run against it. There is no per-team off switch, because a check a workspace can quietly disable is not a check the product can promise.

One repair attempt on a build that failed

A patch that actually failed to build is re-drafted with the failing log in hand, on its own budget, then re-verified. A repair that does not work changes nothing: the original patch, verdict and trace stand.

The ceiling nothing turns off

A pull request is the most that happens unattended. Five things decide whether it lands ready for review or as a draft carrying the reason it was downgraded, and no path in the codebase merges.