Draft with AI
Draft with AI is the fastest way to build a spotlight: describe what you want to see, and the AI fills the builder with a reviewable draft.
Where it runs
Section titled “Where it runs”Spotlight uses Atlassian-hosted AI (the Forge LLMs platform). Your prompt is processed inside the Atlassian platform - it is never sent to any third-party service, and Bytera never sees it. This is the same principle as the rest of the app: nothing leaves your Atlassian tenant’s trust boundary.
What the AI sees
Section titled “What the AI sees”When you click Draft with AI, Spotlight sends:
- Your description - the text you typed.
- Light site context - your site’s project keys and names, and its status names. This is what lets the AI write JQL that’s valid for your site (real project keys, real statuses).
Spotlight does not send issue data, summaries, comments, user lists, or anything else. The AI drafts a definition; the data is fetched later, in your browser, with your own permissions.
What comes back
Section titled “What comes back”A draft that fills the builder form, in one of four shapes:
| Your request looks like… | The AI drafts… |
|---|---|
| A breakdown of issues (“open bugs by assignee”) | An issue chart - JQL + chart type + grouping + metric |
| A sprint question (“velocity for the last 6 sprints”) | A sprint velocity chart |
| A permissions question (“who has access to which projects?”) | A project access table |
| Anything beyond the fixed sources (“SLA performance this month”, “10 oldest unresolved issues with days open”) | A custom plan - a read-only pipeline it composes for your request |
The draft is never saved automatically. You review the form, adjust anything, and save yourself.
Honesty built in
Section titled “Honesty built in”- Caveats. If the draft is an approximation - the request asked for something Jira data can’t express exactly - Spotlight surfaces the caveat in a warning above the form, so you know the difference before you save.
- Out of scope. If the request needs data or actions Spotlight can’t reach (e.g. sending emails, writing to Jira), it says so plainly instead of drafting something misleading.
- Validation. Drafted JQL is checked against Jira’s own parser, and the query is dry-run against the search API before saving - a spotlight that can never render is never saved.
Fair use
Section titled “Fair use”AI drafting is included with the app - there’s no token billing or separate AI subscription. To keep it that way, drafting has monthly fair-use caps:
- Per user: 150 drafts per month.
- Per site: a shared budget of 1,000 AI calls per month - a draft normally uses one call, occasionally two when Spotlight automatically retries an invalid draft - plus an overall token budget.
Counters reset monthly. When a cap is reached, Spotlight tells you which one (yours or the site’s), and the manual builder and templates keep working as usual - the caps apply only to AI drafting.
Tips for better drafts
Section titled “Tips for better drafts”- Name the project: “in project PAY” beats “in our main project”.
- Name the shape: “as a pie chart”, “as a table”, “per month”.
- One question per spotlight. Two questions make two good spotlights.
- Issue Charts - the vocabulary the AI drafts into.
- Custom Plans - what happens when your request goes beyond charts.