Ai Content Bundle Laravel Package

Getting Started

Install the bundle with composer require xmon-org/ai-content-bundle, then publish its configuration (php artisan vendor:publish --tag=ai-content-config) and set your provider credentials in .env (e.g. AI_CONTENT_API_KEY). The bundle is originally a Symfony bundle, so in a Laravel app expect to wire up the service container bindings and routes yourself — check config/ai-content.php and the package's README first, since that is where the provider, model, and storage defaults live.

Your first meaningful use case: generate a draft record (article body, SEO description, or product copy) from a prompt, then preview it in the bundled admin UI. If the generated content includes an image, the admin lightbox is the quickest way to sanity-check output before it reaches users. As of 1.12.0, the lightbox now displays the original image rather than a resized derivative — useful when judging detail, sharpness, or text legibility in generated images.

Implementation Patterns

• Service-first generation. Resolve the content generator from the container and call it from a controller or queued job rather than embedding provider SDK calls directly. This keeps you swappable across providers and lets the bundle handle retries.
• Queue it. AI calls are slow and flaky. Dispatch a job that generates content, writes it to your model, and marks a status flag (draft/ready). Poll or broadcast for completion instead of blocking the request.
• Draft → review → publish. Treat generated content as drafts. Store it on a nullable field and gate promotion behind an approval step; this is where the admin lightbox (with the 1.12.0 original-image view) fits in the review workflow.
• Image handling. The grid/listing typically shows a thumbnail for performance; the lightbox loads the original. If you store both variants, keep the original reference on the record so the lightbox can resolve it — this is exactly the path 1.12.0 exercises.
• Prompt templating. Keep prompts in config or a dedicated class rather than inline strings so you can version them and A/B test output quality.
• Customization via events/listeners. Hook post-generation events to run moderation, translation, or formatting before persisting.

Gotchas and Tips

• Original-image lightbox = heavier payloads (1.12.0). The feature is a real quality win, but the lightbox now fetches the full-size asset. Large originals will be slow and memory-hungry. Ensure originals are reasonably sized, serve them from a CDN, and avoid loading originals for grid previews.
• Symfony heritage. Being a "bundle," some pieces (DI config, event dispatchers, form types) are Symfony-shaped. Verify each integration point against your Laravel container; don't assume Laravel conventions apply.
• Timeouts and rate limits. Set generous HTTP timeouts and backoff on your queue. Provider rate limits are the most common day-to-day failure.
• Determinism. Same prompt rarely equals same output. Snapshot-test structure, not exact text.
• Cost visibility. Log token/usage per generation so you can attribute spend per feature.
• Config caching. After changing provider/model config, run php artisan config:clear — stale cached config is a frequent "why is it still using the old model" culprit.
• Debugging. Log raw provider responses on failure; the bundle's exceptions often wrap the real error. Check the admin UI's rendered output (and lightbox) as a fast end-to-end check that generation and storage both worked.
• Extension points. Prefer the bundle's events and config overrides over forking; the 1.12.0 UI tweak landed cleanly precisely because it was additive, but future breaking changes are less likely to account for local forks.