Getting started
From install to your first evidence-backed findings in a single admin session — under 30 minutes.
Before you start
- You need a Confluence Cloud admin role to install the app and select spaces to scan.
- 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 on the Marketplace listing.
- No setup, no keys. There are no API keys to create and no external services to connect — the app runs entirely on Atlassian’s Forge platform.
1. Install from the Atlassian Marketplace
Find Evergreen AI for Confluence on the Atlassian Marketplace and choose Get it now (or Try it free). Installation is self-serve and takes a minute.
Because the app uses Atlassian-hosted AI (the Forge LLMs capability), the install prompt asks you to approve the app’s access. Here is exactly what you are approving:
| The app can… | What that means |
|---|---|
| Read page and space content | So it can analyze what your pages actually say. Read-only. |
| Use Atlassian-hosted AI | Analysis runs on Atlassian-hosted models inside Atlassian’s infrastructure. Nothing is sent to any vendor-operated or non-Atlassian service by this app. |
| Store its own data | Findings, assignments, and scan history live in Atlassian-hosted Forge storage. |
| Write a page comment (optional, off by default) | Only if you turn it on later: a single comment that @mentions an assignee, so Confluence sends its own notification. See Settings. |
What the app never does: it does not modify, archive, or delete the body of any page, and it makes no external network calls of any kind. See Security & data handling for the full picture.
2. Open Evergreen AI and run the onboarding wizard
After install, open Evergreen AI from the Confluence app menu. The first time you do, a three-step wizard replaces the dashboard until your first scan completes.
- Pick one space to audit. Choose a single space to start with. Nothing is scanned until you select it — the app never touches a space you haven’t chosen.
- Run the sample scan. The app scans up to 100 pages with the Standard model, regardless of how large the space is. You’ll see a live progress bar and findings appearing as they’re confirmed. This usually takes a few minutes and is included in your trial — no surprise cost.
- Review your first findings. The wizard frames the result honestly: for example, “We found 4 things worth a look — and 89 pages that are perfectly fine.” From here you can review the findings or schedule recurring scans.
If the sample comes back with zero findings, that’s a real result, not a failure. The app will suggest pointing it at a messier space next — runbooks and policies are usually where problems hide.
Prefer to configure everything first? Admins can dismiss the wizard and go straight to Settings to choose scope, model tiers, schedule, and sensitivity before running anything.
3. Choosing a good first space
The sample scan is capped at 100 pages, so pick a space where problems are likely to be concentrated:
- Engineering runbooks — deployment steps, escalation contacts, and tool references go stale fast.
- Policies and procedures — documents that often declare their own expiry (“valid until…”).
- Onboarding and IT how-tos — full of system names and ownership that drift over time.
Once you trust the results, widen the scope and set a schedule — see Scans & scheduling.
What good looks like
Within your first session you should be able to: open the dashboard, see which space is worst, read the evidence for the top finding, and assign it to an owner — in a handful of clicks. If you can, the app is doing its job. Next, learn how to read what it surfaces in Understanding findings.