An AI assistant that actually knows retail.

What it can actually do

Most "AI for retail" pitches stop at "ask anything." Vendee Pro’s assistant can do that, and it can also act. It has direct, audit-logged access to a tool catalog covering purchase orders, inventory, vendors, products, categories, and open-to-buy. Ask it which vendors ship slowest, and it queries your PO history. Ask it what’s under-stocked across a category, and it returns the list. Ask it to draft a PO, and it proposes one, line by line, for you to confirm.

Confirm before write

Anything that mutates your data, creating a PO, editing a product, adjusting inventory, has to be confirmed by you. The assistant proposes the change, shows the preview, and waits. You either click confirm or you don’t. There is no path for the AI to change your records without an explicit human check. This is not a marketing claim. It is the architecture: write tools are gated by a confirmation layer that the model cannot bypass.

That single design decision is what makes the assistant safe to use on real data. The work the model does is reversible by simply not confirming. The work you do is auditable, because every confirmed write is logged with the prompt, the proposal, and the user who approved it.

Tenant-isolated and audit-logged

Every read the assistant performs is scoped to your account at the data layer through multiple independent layers of isolation: immutable server-built context, function-calling-only (no raw SQL), row-level boundary assertion on every returned row, forensic audit log for every tool invocation, and account-scoped conversation history. Every tool call is logged (which tool, what arguments, rows returned, duration, status) so you can review the assistant’s activity the same way you review a buyer’s.

Voice that fits the floor

On a trade-show floor, on a stockroom shelf, in a back office where typing isn’t the fastest input, the assistant accepts voice. Speech is transcribed by your device or your browser’s native speech recognition, then sent through as text. We don’t record continuous audio. The transcript becomes part of the conversation, the assistant proposes its action, and you confirm by tapping or by saying yes.

Designed for the way buyers actually think

You don’t think in SQL. You think in questions: "What did we buy from Vendor X last spring?" "Are we behind on this category against OTB?" "Which sizes are we running out of in the green tee?" The assistant is shaped around questions like that. It maps your phrasing to the right tool, runs the query, and returns an answer in the same plain English you asked in. When the answer would be better as a chart or a table, it returns the chart or the table.

And when a question would result in a write action, the assistant pivots: "Want me to draft a PO for that?" Click yes, review, confirm. The flow stays a conversation.

What it deliberately does not do

The assistant is not a replacement for the buyer. It does not auto-publish to your store. It does not place orders with vendors on its own. It does not move money. It is a high-context co-pilot, not an autonomous agent. That boundary is where retail’s real risk lives, and we keep the human on the right side of it.

Frequently asked questions