Automations
An automation is a rule that does a piece of triage for you. Every rule reads as one sentence: When something happens, If the report matches, Then do this. When a report arrives, if the AI category is Bug and it has console errors, then create a ticket in Jira. Nobody has to click.
Open Automations in the space navigation. The page has two tabs: Rules, the rules in the order they run, and Templates, six starting points you can add with one click.
When: the triggers
Section titled “When: the triggers”A rule starts on exactly one of nine moments:
| Trigger | Fires when |
|---|---|
| A report arrives | A new report is filed from the widget, the extension or the API. |
| A report is updated | Its status, priority, assignee or labels change. |
| A comment is added | Someone writes on the report’s internal thread. |
| A pin is resolved | A pin on the screenshot is marked done. |
| A duplicate is detected | A report is found to be the same problem as an open one. |
| A cluster grows | A group of similar reports passes a size you set. |
| A spike is detected | Volume jumps to a multiple of what is normal. |
| A report sits untouched | A report has had no attention for the hours you set. Checked every 15 minutes. |
| A survey is answered | Someone answers one of your surveys. This trigger carries no report, so actions that need one are refused when you save. |
If: the conditions
Section titled “If: the conditions”Add as many conditions as you need (up to 20), and choose whether all of them or any of them must match. A rule with no conditions acts on every report its trigger fires for.
- Where it came from: project, page address (contains, does not contain, equals, or matches a
wildcard pattern where
*means anything), device, browser, capture mode. - What it carries: has console errors, has failed requests.
- Its triage state: label, priority, status.
- Who sent it: reporter type (tester, site visitor, API, survey), reporter email domain, and the report’s text.
- The change itself (for a report is updated): which field changed and its new value.
- Trigger-specific numbers: hours untouched, cluster size, spike size (× normal), survey score.
And the conditions that read the AI analysis:
| AI condition | Values |
|---|---|
| AI category | Bug · UX / design · Content / copy · Performance · Feature request · Question · Praise |
| AI severity | Critical · High · Medium · Low |
| AI area | The component or page section the report is about, as free text |
| AI sentiment | Negative · Neutral · Positive |
| AI confidence | How sure the model is of the category, from 0 to 1 (for example at least 0.8) |
| AI severity confidence | How sure it is of the severity |
| AI duplicate match | How closely the report matches an open one, from 0 to 1 |
When a report has no analysis (AI is off for the account, out of credits, or the call failed), every AI condition counts as not met, and the run log says why. A rule never guesses.
Then: the actions
Section titled “Then: the actions”A rule can run up to ten actions, in order. There are fifteen:
| Action | What it does | Undo |
|---|---|---|
| Add a label | A label you pick, or the label matching the AI category. | Yes |
| Set priority | A priority you pick, or the priority matching the AI severity. | Yes |
| Set status | Moves the report to a status. | Yes |
| Assign to | A teammate you pick. | Yes |
| Assign to the area’s owner | Whoever has handled reports of the same area in this project before — counted from your history, not chosen by the model. | Yes |
| Assign in turn among | Round-robin among the people you list. | Yes |
| Merge into the detected duplicate | Merges the report into the open one the analysis matched, above a minimum match. | Yes |
| Create a ticket in the tracker | Files an issue in the project’s connected tracker, with AI-written steps to reproduce. | No — it has left uMarker |
| Post to a channel | Posts to one channel, or to every channel attached to the project. | No |
| Notify a person | Sends a teammate a notification. | No |
| Email the reporter | “We fixed it” or “this is already being tracked”, in the reporter’s language. Skipped if they have unsubscribed or left no verified email. | No |
| Add to the changelog draft | Adds a resolved report to the changelog draft. | Yes |
| Set a due date | A deadline a number of hours away. | Yes |
| Escalate | Bumps the priority and notifies. | Yes |
| Run the AI analysis | Asks for the analysis on a trigger that does not get one by itself. | No |
Actions that send something out of the product cannot be taken back, so they have no Undo.
AI analysis on every new report
Section titled “AI analysis on every new report”When automations are on your plan, every new report is read once by the model, which produces a structured verdict: category, severity, area, sentiment, language, a clean title, a developer-ready ticket (steps to reproduce, expected and actual), a likely cause from the console and network when the report is Technical, duplicate candidates with a score, and a suggested owner. It is stored on the report as facts the rules act on, not as a suggestion waiting for a click.
It goes through exactly the same gates as every other AI feature in uMarker — see AI triage assistant:
- Plan: Professional and above.
- Consent: your account has switched AI on, and the report arrived after that. A report filed before consent is never sent.
- Credits: each analysis costs 4 credits from the month’s allowance. With your own provider key, your provider bills you instead.
- Your switch: Report analysis has its own switch on your AI settings page.
What leaves is the report’s text only (the comment, the page URL, the redacted console) — never the screenshot, the email address or the network log. On any other trigger, the analysis runs only when one of that trigger’s rules actually reads it.
Templates
Section titled “Templates”Six one-click starters. Each adds an ordinary rule, switched off, that you can read and change before it does anything:
| Template | What it does |
|---|---|
| Auto-triage | Adds the AI category label and sets priority from the AI severity, when the AI confidence is at least 80%. |
| Merge obvious duplicates | Merges a report into an open one the AI reads as the same problem (match ≥ 0.9), and tells the reporter it is already tracked. |
| Bugs with console errors → tracker | Files a ticket with AI-written repro steps for every Bug that arrived with console errors. |
| Critical → channel + owner | Posts critical or urgent reports to the team channel right away and assigns the area’s owner. |
| Close the loop | When a report is resolved, emails the reporter in their language that it is fixed. |
| SLA: escalate the neglected | Escalates a new report nobody has touched for 24 hours. |
Build and try a rule safely
Section titled “Build and try a rule safely”-
Press New automation (or Use template). Pick where it applies — every space, this space, or one project — then fill in When, If and Then. Below the builder, the rule reads back as one sentence, so you can check it says what you meant.
-
Switch on Dry run. The rule then runs for real but changes nothing: the run log records what it would have done.
-
After saving, use Test on a report: pick a report and see whether the rule would act on it and what it would do. Nothing is changed.
-
Read the Run log — the last runs, each with what matched, what was done, what failed and why. When you are happy, switch dry run off and the rule on.
Rules run top to bottom. Drag a rule to reorder it, or focus its handle and use the arrow keys.
Everything the machine did is visible and undoable
Section titled “Everything the machine did is visible and undoable”On the report page, Done by automations lists every automated action with the rule’s name, the AI confidence and the reason — for example Labelled Bug — 92% sure. Each one that can be reversed has an Undo button that puts the previous value back.
If somebody has changed that value since the automation did, Undo refuses and says so: undoing would overwrite a person’s later decision with an older one.
An undo is also a correction the AI learns from: an undone AI label or priority is remembered the same way as a manual relabel, so the next analysis in this space is better (unless you have switched off learning from your team’s corrections on the AI settings page).
The Automations page also shows, per rule: runs, actions in the last 30 days, the undo rate and the last run. And the space home carries a card: Automations handled X% of triage this week, with the number of actions and how many were undone — counted, not estimated.