FAQ & troubleshooting

Frequently asked questions

Does Evergreen AI edit, archive, or delete my pages?

No. The app is read-only on your page content — it never modifies, archives, or deletes the body of a page. Its only writes are its own data (findings, assignments, scan history) and, if an admin turns it on, a single comment that @mentions an assignee. It does not auto-fix or rewrite pages.

Where does my content go? Does anything leave Atlassian?

Nothing leaves Atlassian. Page content is processed by Atlassian-hosted models (the Forge LLMs capability) inside Atlassian’s infrastructure, and is never sent to any vendor-operated or non-Atlassian service by this app. The app declares no egress permissions and makes no external network calls of any kind. It’s designed to qualify for the Runs on Atlassian badge. For specifics on how Atlassian handles AI inference on its platform, see Atlassian’s own Forge platform and AI terms; for the app’s data inventory, see Privacy and Security & data handling.

Which AI model does it use?

Two tiers, both Atlassian-hosted Claude models via Forge LLMs: Standard (Claude Haiku) is the cost-efficient default, and Deep (Claude Sonnet) is for high-stakes spaces. You set the tier per space in Settings.

How is this different from timestamp-based lifecycle tools?

Those tools flag pages for being old — by last-edited date, view count, or a fixed review interval. That misses the page edited yesterday that still says “deploy with Jenkins,” and nags about evergreen pages that are perfectly fine. Evergreen AI reads what a page actually says and surfaces the ones most likely to be wrong, with a quote and a reason for each.

Couldn’t I just build this with a Rovo agent?

Evergreen AI’s value isn’t a single model call — it’s the system around it: bulk scanning across thousands of pages, a precision layer that suppresses false positives, a suppression memory that learns from your dismissals, an evidence-first dashboard, and a scan ledger. That’s what turns “ask the AI” into a repeatable, defensible audit.

What does it cost?

Free for teams of ten or fewer. Larger teams are billed per user through the Atlassian Marketplace, with a 30-day free trial. Exact pricing is shown on the Marketplace listing. AI usage is covered by the app — there are no separate AI bills or API keys for you to manage.

Will it flood me with alerts?

No — that’s the whole point. Below its confidence threshold the app says nothing, it caps findings per page, and once you dismiss something with a reason it won’t raise it again. A clean space simply shows “all clear.”

Can dashboard viewers see content they don’t have access to?

No. Before showing a finding’s evidence, the app checks whether you can view the underlying page. If you can’t, the row is redacted to its type and space only — no title, quote, or link.

How do I remove all the app’s data?

Use Purge all app data in Settings → Data controls for an immediate, irreversible wipe. Uninstalling the app also removes its data per Forge’s standard behavior.

Troubleshooting

“Scanning paused — monthly budget reached”

You’ve hit the server-side monthly budget. Scanning resumes next cycle automatically; triage and the dashboard remain fully available in the meantime. You can review usage in Settings → Usage & budget. This is a deliberate guardrail — the app caps spend rather than running it up.

A scan only covered part of a space

A scan can come back partial if it didn’t finish within its window or paused at budget. The app publishes what it completed and resumes in the next scheduled window. The Overview banner shows the coverage percentage, and the Scans tab shows planned-versus-completed pages. Very large instances run in rolling-coverage mode by design — see Scans & scheduling.

A page shows “Not yet scanned”

Its space probably isn’t in scope yet, or it hasn’t been reached by a scan. Add the space in Settings → Scope and run Scan now. Note that the app skips very short pages, archived pages, and templates by design.

A finding says “analysis failed” for a page

Occasionally a page can’t be analyzed — for example, an extremely long page, or one the platform’s content moderation declined. The app records this in the scan ledger and moves on rather than wedging the run. Extremely long pages may be truncated for analysis, which is noted on any resulting finding. You can re-run the scan later.

A finding row says “a finding exists on a page you can’t view”

That’s the permission-safe behavior working as intended: you don’t have view access to that page, so its evidence is hidden. Ask the page’s space admin for access, or have someone with access triage it.

My first scan found nothing

That’s a valid result — the sampled pages held up. Try pointing the app at a messier space; runbooks, policies, and IT how-tos are where problems usually hide. You can also raise sensitivity to Thorough in Settings to cast a wider net.