Skip to main content

AI Catalog Wizard

The AI Catalog Wizard reads an unstructured source document — a PDF policy guide, a Word-format control list, an HTML page, or plain text — and drafts a structured OSCAL Catalog you can review and save. Instead of hand-coding XML or JSON, you supply the source material and the wizard handles extraction, structuring, and initial validation.

What it does

The wizard processes your input in three stages:

  1. Extraction — The AI reads the source and identifies control-like constructs: titles, IDs, requirement statements, parameters, and groupings.
  2. Structuring — Extracted controls are organised into the OSCAL catalog model: groups, controls, and parameters with proper identifiers and metadata.
  3. Handoff — The draft is passed directly to the Catalog Builder so you can review, edit, and save without leaving the application.

The wizard targets the OSCAL Catalog model. If your goal is a Profile (a selection from an existing catalog) or a Component Definition (control implementations), use the appropriate wizard or builder instead.

Supported inputs

Input typeNotes
PDFText-based PDFs only; scanned images may extract poorly
Word (.docx)Formatting and headings are used to infer structure
HTMLPage text is extracted; navigation and boilerplate are stripped
Plain textPaste directly or upload a .txt file

Maximum upload size is approximately 10 MB. Very large PDFs are processed in chunks, which may increase generation time.

How to use it

  1. Open the AI Wizard hub

    Navigate to /ai/wizard or click AI Wizard in the main navigation. The hub displays a card for each available wizard.

  2. Select 'Build Catalog from Source'

    Click the Build Catalog from Source card, then click Start →. The wizard opens to the upload step.

  3. Upload your source or paste text

    Either drag-and-drop a PDF, Word, or HTML file onto the upload area, or click Browse to select a file. Alternatively, switch to the Paste text tab and paste your source content directly. Click Next when ready.

  4. Generate the draft

    Click Generate. A progress indicator shows the three stages: Extraction → Structuring → Validation. Generation typically takes 15–90 seconds depending on document size. The wizard retries automatically if the Anthropic API is temporarily busy.

  5. Review the draft in the Catalog Builder

    When generation completes, the wizard navigates you to the Catalog Builder with the draft pre-populated. Review every group, control, parameter, and piece of metadata. The AI sometimes invents plausible-sounding but incorrect content — treat this as a starting point, not a finished document.

  6. Edit, validate, and save

    Make any corrections in the builder. When satisfied, run schema validation on the Review & Save step to confirm the document is valid OSCAL, then click Save to Library to publish it.

Tips and limits

  • Large PDFs take longer. A 200-page PDF may take several minutes to process. Keep the tab open and watch the progress indicator.
  • Review everything. The AI sometimes merges separate controls, splits a single control into multiple entries, or generates IDs that look plausible but differ from the source. Always compare the draft against the original document.
  • Re-run with a better source. If the first draft is poor quality, try editing the source text to make structure more explicit (add headings, remove irrelevant boilerplate), then generate again.
  • Scanned PDFs extract poorly. Use text-based PDFs where possible. If you only have a scanned document, convert it with an OCR tool first.
  • AI features must be enabled. If the wizard card is greyed out or you see "AI features are disabled," ask your Org Admin to enable AI at /org-admin/ai-settings.

The draft is passed to the Catalog Builder using session storage. If you refresh the browser between clicking Generate and arriving at the builder, the draft will be lost and you will need to generate again.

Related