Investigate a bug report with missing details

How Novix gets involved

The original customer thread.

The question goes back on the thread the report arrived on. Answering is replying, in the tool the reporter is already in, with nothing to learn and nothing to open.

How the agent handles it

Novix identifies what is missing
ClassifyDecides whether this is actionable work on your repo and sets its category and severity, setting spam and outside feedback aside without spending anything.
A technical report with no error text, no page or feature named and no steps sets a needs-info flag, plus the specific details worth asking for, capped at 300 characters.
The investigation waits for evidence
DiagnoseThe root-cause investigation: it opens the files it suspects in your repo, then names the cause, a confidence and the areas it affects.
The deep investigation is skipped outright and the verdict is written with no model call at all. A report nobody could begin to diagnose never reaches the expensive lane.
One question, not a questionnaire
ClassifyDecides whether this is actionable work on your repo and sets its category and severity, setting spam and outside feedback aside without spending anything.
The verdict names exactly those missing details, so what goes back on the thread asks for the thing that is actually blocking the run rather than a generic list.
The answer restarts the run, once
A follow-up on the thread re-runs the pipeline when the verdict is needs-info, never on Novix’s own comment coming back through the webhook, and at most once per question asked.
A colleague gets searched for instead of asked
When the reporter is a member of your workspace and there is a repo connected, the thin report goes to the investigation anyway. Novix has something they do not, which is a search over their own code.

What your team gets

The one question that would settle it, on the thread it arrived on
A re-run the moment somebody answers, without a dashboard visit
For a colleague, what was already searched rather than an open question

A screenshot is never treated as a thin report

A ticket carrying an attachment never takes this path, because the picture is often the whole report and the diagnosis stage is the one that gets to look at it.

Works with

Zendesk
Intercom
Slack
Linear

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.

How the agent handles it

Novix identifies what is missing

A technical report with no error text, no page or feature named and no steps sets a needs-info flag, plus the specific details worth asking for, capped at 300 characters.

The investigation waits for evidence

The deep investigation is skipped outright and the verdict is written with no model call at all. A report nobody could begin to diagnose never reaches the expensive lane.

One question, not a questionnaire

The verdict names exactly those missing details, so what goes back on the thread asks for the thing that is actually blocking the run rather than a generic list.

The answer restarts the run, once

A follow-up on the thread re-runs the pipeline when the verdict is needs-info, never on Novix’s own comment coming back through the webhook, and at most once per question asked.

A colleague gets searched for instead of asked

When the reporter is a member of your workspace and there is a repo connected, the thin report goes to the investigation anyway. Novix has something they do not, which is a search over their own code.