Generation Reliability & Media Support
A full reliability pass on the generation engine — output quality guardrails, faster file handling, three production bug fixes, and Google Ads tracking.
- AILexicon quality guard — generated campaigns are now validated against a banned-phrase list (AI clichés, filler structures, weak closers). Flagged drafts are automatically repaired with a second targeted call before being returned to the user.
- ImprovementFaster file & image generation — images, audio, and PDFs are now passed to Vertex AI as direct HTTPS URIs instead of being fetched and base64-encoded, cutting per-file overhead and keeping requests well within the 60s runtime limit.
- ImprovementFile upload warning — an amber notice now appears whenever attachments are added, advising users that larger files increase generation time.
- FixComposio OAuth redirect — connecting integrations no longer redirects to localhost in production. The callback URL is now derived from the live request origin.
- FixPersona manager infinite spinner — the loading state now always resolves, even when the Supabase fetch errors or returns no data.
- FixGeneration stuck at loading — restored the synchronous generation flow that was silently failing in production due to a misconfigured async worker pipeline.
- InfraGoogle Ads conversion tracking (AW-18111303438) wired up across the main app and blog via a shared `gtag` module.
