Replicate for WP hands the input to a model hosted on Replicate and drops the response back into WordPress as a post or an attachment. Replicate owns the inference. WordPress owns the artifact: a row in wp_posts with title, status and author, plus wp_postmeta keys recording the model version slug, the prediction ID and the prediction status that produced it.

That artifact is what an editorial or ops team can actually triage, and what the default Posts screen handles poorly. SleekView reads the same wp_posts rows and the same meta directly. Title, status and author sit alongside model version, prediction ID and prediction status as real columns. Sort by date, filter to predictions on a single model version, or pull every run that returned failed, all without opening each post.

Edits run through standard WordPress CRUD, so save_post hooks still fire and any Replicate-side meta the plugin reads on update stays consistent. Bulk-flip queued drafts on a single model to pending review in one pass.

Replicate for WP makes calling a hosted model cheap enough that prediction volume rises before any reporting is in place. The default Posts screen turns that volume into a wall of titles with the model version and prediction status hidden a click away. SleekView reads the same wp_posts rows and the same meta and turns them into columns a team can sort, filter and edit.

Ops stops opening every post to check whether a prediction succeeded. Content leads stop guessing which model produced which draft. Governance stops second-guessing disclosure coverage.

Same data, very different conversation in the weekly engineering review.