Hoobert - AI Command Bar for WooCommerce

No. Hoobert does not use a general-purpose LLM. It talks to one Fernfly inference endpoint running a small model fine-tuned on Hoobert's own WooCommerce actions, and that is the only outside service involved.

The plugin is free and GPL, forever. The inference endpoint is a Fernfly project, which has a free tier that is enough for normal single-store use. There is no locked feature, no trial timer, and no upsell inside the plugin.

It is constrained three ways at once. It can only run one of 28 predefined WooCommerce actions; it runs them under your own admin capabilities, so it cannot exceed what you could do by hand; and anything destructive requires an explicit confirm click after you have seen exactly what will happen. Everything it does run is logged. That said, a refund is a refund. Read the confirmation before you click it, the same as you would in WooCommerce itself.

Only the sentence you type, plus the order or product id of the screen you are on. Nothing is sent unless you run a command. Customer data, order contents, and your credentials never reach the inference endpoint. The model only needs your words to decide which action you meant; the action itself runs locally.

Only logged-in users with the manage_woocommerce capability, meaning shop managers and administrators. Hoobert's REST routes enforce this on every request, not just in the UI.

Yes. Hoobert lives entirely in wp-admin and does not touch your storefront, your theme, or your front-end at all. Visitors never load a byte of it.

Both the inference endpoint URL and the API key must be filled in under WooCommerce -> Hoobert. If both are set and you still see it, check that the URL is your project's full infer route (it ends in /infer) and that the key was copied without surrounding whitespace.

Rephrase closer to the action you want: "refund order 1042" beats "sort out that customer's money". If a request is outside the 28 actions, Hoobert says so rather than guessing. But if you genuinely believe Hoobert got it wrong, head over to Fernfly > Your project > Retrain, find the utterance that it got wrong, correct the function-call JSON in the textbox, save that utterance-function pair, and click the Retrain button. After your model has finished training, Hoobert will start behaving as expected.

Yes. The tool set is tools.json in the plugin folder: an OpenAI function schema per action, plus a small block describing the REST call to dispatch. Add your entry there, register the matching schema with your Fernfly project, and retrain. The repository README walks through it.

WooCommerce -> Hoobert, below the settings, shows every command every admin has run on this store, newest first. The palette itself also has a Hoobert: Query history command showing your own recent ones.

The interface is translatable. What the model understands depends on the Fernfly project you train, so a non-English store can train on non-English utterances.

Yes. Settings and command history are per-site, so each store in a network has its own endpoint and its own audit log.

Yes. The repository has a WordPress Playground blueprint that spins up a throwaway store in your browser, with WooCommerce, Hoobert, and sample data already set up.