Settings & budget
Everything an admin controls — scope, schedule, sensitivity, suppression, the monthly budget, and your data.
Settings is a tab within the Evergreen AI dashboard. Each section saves on its own, so you can change one thing without committing to the rest.
Scope
Choose which spaces are in scope and set each one’s model tier — Standard (Claude Haiku) or Deep (Claude Sonnet). Each space shows an estimated page count so you can gauge the impact before you commit. Remember: a space is only ever scanned after you add it here.
Schedule
Set each scope to run Off, Weekly, or Monthly, and trigger a one-off Scan now. See Scans & scheduling for how recurring, incremental scans behave.
Sensitivity
Choose how confident the app must be before it shows a finding:
| Setting | Confidence threshold | Effect |
|---|---|---|
| Conservative | 80 | Only the most certain findings. |
| Default | 70 | The recommended balance. |
| Thorough | 60 | A wider net, more to review. |
You can also toggle individual finding types on or off here. Cross-page contradiction detection carries a “beta” label while it’s being calibrated — see Understanding findings.
Suppression list
Every finding you dismissed with “reduce future flags like this” is collected here, so the app’s learned suppressions are always reviewable — and removable. If you suppressed something by mistake, take it off the list and it can surface again on the next scan. This is the human-readable record behind the suppression learning loop.
Usage & budget
Evergreen AI runs on a server-side monthly budget, enforced in code. You can’t accidentally configure your way into unbounded AI spend — the budget is a hard wall, not a soft suggestion. This section shows the month’s usage against the cap, an estimate of pages remaining, and a short history, with copy explaining how the cap is derived.
| At… | What happens |
|---|---|
| 80% of budget | You get a warning, so there are no surprises. |
| 100% of budget | Scanning pauses with clear messaging. Triage and the dashboard keep working fully — you can still confirm, dismiss, assign, and resolve. Scanning resumes next cycle. |
The guiding rule: when the app runs low on budget, it scans fewer pages and tells you — it never spends more. Behind the scenes it also respects Atlassian’s rate limits and checkpoints its progress, so an interrupted scan resumes without re-spending on pages it already analyzed.
Assignment notifications (optional)
By default, assignments appear on the page panel and in the assignee’s “Assigned to me” filter — no extra permissions needed. If you want assignees to get a native Confluence notification, you can enable the optional comment-mention: the app posts a single comment that @mentions the assignee on the affected page, and Confluence handles the notification. This is off by default and is the only feature that writes anything to a page — and even then, it’s a comment, never an edit to the page body.
Data controls
You’re always in control of the app’s data:
- Purge all app data — an immediate, irreversible wipe of everything the app stores (findings, assignments, scan history, suppression list). It’s guarded by a typed-confirmation modal so it can’t happen by accident.
- Uninstalling the app also removes its data per the Forge platform’s standard behavior.
All app data lives in Atlassian-hosted Forge storage. For the full data inventory and deletion details, see Privacy and Security & data handling.
About
The About section shows the app version, the current prompt version, the “designed to qualify for Runs on Atlassian” statement, and a link to support.