Skip to main content

Aura Engine

The Aura Engine takes the output of an extraction and rebuilds it as a working site inside the Aura runtime — AURAFUSE's editable, publishable website format. Where AuraExtract studies the original, Aura Engine recreates it: structure, styling, and content, as a real codebase you can edit.

A migrated brand open in the Aura studio editor

One-click migration​

Open a brand with a completed extraction and start the migration. Aura Engine runs the rebuild as a background job while you watch a stage checklist update in real time — the same pattern as extraction, with per-stage pass/fail states and status.

LLM-driven rebuild with self-correction​

The rebuild is not a mechanical template swap. Aura Engine drives an LLM-powered rebuild loop:

  • The engine feeds the extracted tokens, layout, and components to the model, which generates Aura replica files — components, theme, pages, and styles.
  • Each stage runs through self-correction loops: the engine validates what the model produced, feeds failures back, and the model retries with the error in context until the stage converges.
  • The result is checked against the extracted design, not just for syntax, so visual fidelity is the success criterion.

This matters because cloned sites are messy in the wild. A self-correcting pipeline absorbs the ambiguity — slightly different button styles between pages, inconsistent spacing — instead of failing on it.

Per-stage retry​

Like extraction, a failed stage does not doom the migration. The stage checklist shows exactly where the job stopped, and you can retry that stage alone. The engine keeps everything that already succeeded and resumes from the failure point, so a flaky stage costs seconds, not a full rebuild.

After migration​

When the migration completes, your brand owns a full Aura replica:

  • Open it in the Editor Studio to edit files, theme, and pages.
  • Links between the original site's pages are remapped to local slugs, so a multi-page clone navigates internally instead of pointing back at the original site.
  • Validate the replica against the original and let auto-fix reconcile gaps.
  • Replace placeholder copy through the CMS.
  • Publish when the site is ready for production.

Tips​

  • Start migrations from a clean extraction — retried-and-fixed token sets migrate more reliably than partial ones.
  • Watch the checklist during the run; the stage where self-correction spins tells you where the original site was hardest to reproduce.
  • If fidelity feels off, fix the design tokens first and re-migrate before hand-editing the replica.