Skip to main content

AI Diagnosis

A page for generating and reading comprehensive diagnostic reports that analyze your account's overall operational state from AWS-native data.

AI Diagnosis report

Features

Run a diagnosis

  • Diagnosis depth (Tier): choose the scope of analysis.
    • Light: a fast check of core items only (currently identical catalog/token budget to Mid)
    • Mid: a balanced standard diagnosis (default)
    • Deep: a broad analysis covering 15+1 sections (16 rendered). Only Deep lets you additionally pick a model (Sonnet / Opus), and choosing Opus shows an accompanying cost note.
  • Report language: choose the report body language (한국어 / English / 中文 / 日本語). It defaults to the current UI language, and switching only the language within the same hour starts a new report.
  • Press the 진단 실행 (Run diagnosis) button and the report is produced asynchronously by a background worker.

Progress indicator

  • While the report is being produced, a live progress bar shows the data collection → section analysis → assembly stages.
  • The report opens automatically once it is done.

Report list

The left sidebar lists your recent reports:

  • An auto-generated title, the id, depth (Tier), model, status, and creation date (KST)

Report body

  • The body renders as collapsible section cards with severity icons and a sticky table of contents (TOC) sidebar on the right.
  • Use the top buttons to export as MD / DOCX / PDF.

Insight badges

  • An intended-vs-actual / change insights badge row summarizes invariant violations and changes versus the previous report.
  • The Intent (invariant candidates) panel lets you propose, accept, and reject candidates. (Admin-only; read-only for everyone else.)

Title, tags, and delete

  • The report owner or an admin can edit the title inline, add and remove tags, and soft-delete the report.

Scheduled diagnosis & notifications

  • Scheduled diagnosis: besides the cadence (weekly/biweekly/monthly) you can pick a weekday (weekly/biweekly), a day of month 1–28 (monthly), a run hour (KST), and the report language; the next run and last run times are both shown. Unset fields keep the interval-only behavior.
  • Diagnosis mailing list: admins can, in addition to adding/removing subscribers, press Send test to deliver one test email to every confirmed subscriber and verify delivery.

How to use

  1. In the sidebar, click AI Operations > AI Diagnosis.
  2. Select a diagnosis depth (Light / Mid / Deep).
  3. If you chose Deep, optionally specify the model (Sonnet / Opus).
  4. Choose the report language (default: the current UI language).
  5. Press the 진단 실행 (Run diagnosis) button to start generation.
  6. Watch the stages on the progress bar; when it finishes, review the report that opens automatically.
  7. Navigate sections with the table of contents sidebar on the right (section cards collapse/expand), and export as MD / DOCX / PDF from the top.
  8. Edit the title and tags or delete the report if needed (when you have permission).

Tips

Duplicate-run protection

If you run the same conditions (depth, model, and language) again within the hour, the existing report is returned instead of generating a new one, along with a duplicate-run notice.

Permissions

Editing the title, tags, and deletion are available only to the report owner or an admin. Proposing, accepting, and rejecting invariant candidates is admin-only and is shown read-only to other users.

Read-only analysis

AI Diagnosis only observes and analyzes AWS resources — it never modifies them. Reports are for analysis purposes only.

AI analysis tips

After reviewing a report, you can ask follow-up questions from the floating button or the AI Assistant page, for example:

  • "What is the riskiest item in this diagnosis?"
  • "What changed compared to the previous report?"
  • "Explain the cause of the invariant violations shown."
  • AI Assistant - follow-up questions about report results
  • Async Jobs - check the progress of diagnosis generation