Home / Docs / Putting a product catalogue into the agent

Putting a product catalogue into an AnroAgents agent

How to prepare a catalogue, price list or property list so the agent answers well.

A catalogue is the hardest thing to get right in any AI agent, and AnroAgents handles it through the Knowledge Base with a specific preparation step. The rule is simple: one row of your catalogue should become one self-contained passage of text.

Reading the catalogue from your shop's website

If your catalogue is on your own website, the simplest route is to add the website as a whole in the Knowledge Base panel. AnroAgents recognises product pages by the product data shops publish for search engines and reads each one as a product card — name, category, price, availability, brand, colour, material, size, variants, photo captions and link — up to your plan's product limit (50 on Free, 500 on Starter, 5,000 on Pro, 20,000 on Business). On Shopify and WooCommerce shops the catalogue is read through the platform's own product list, which is faster and gentler on the shop than loading every product page. On Pro the shop is read again automatically every week and on Business every day; on other plans press the refresh icon on the source after prices or the range change. A product page that carries no such data is read as an ordinary page; a shop whose host shows bots a security check cannot be read this way, and for it the CSV route below is the answer.

Uploading a catalogue as CSV

The AnroAgents Knowledge Base accepts CSV, and CSV is the best format for a catalogue. Each row is turned into a card that keeps its column labels and is indexed as its own passage, so a row about a product carries its own name, price, availability and description together, and a question naming several attributes at once — a green velvet corner sofa around €2,000 — finds the one row that has all of them.

A CSV with more than 20,000 rows, or with rows shorter than about 100 characters on average (a bare price list of names and prices), is indexed with several rows to a passage instead, so that it indexes in time. Each row still keeps its column labels. To keep one passage per row in a catalogue of more than 20,000 items, split it into several CSV files by category.

Export the catalogue from your shop, CRM or spreadsheet as CSV with a header row, and give the columns names a person would use: name, price, category, availability, description, url. The agent quotes those labels back, so price_eur_incl_vat reads better than p2.

Uploading a catalogue as a Markdown or text file

An AnroAgents agent also reads a catalogue written as text, provided every item starts with its own heading line (## Hoekbank Gardolo) followed by labelled lines such as Category: …, Price: …, Colour: …, Material: …. When a document holds at least 20 such items, each with at least three labelled lines, the AnroAgents Knowledge Base indexes every item as its own passage, so a question naming several attributes at once — a green velvet corner sofa around €2,000 — finds the one item that has all of them. Pages of ordinary text in the same document, such as delivery terms, are indexed as before.

Keep the AnroAgents catalogue CSV to the columns that matter in a conversation. Internal identifiers, warehouse codes and supplier margins add noise, count against the character limit and occasionally leak into an answer.

What to do with a large catalogue

An AnroAgents agent searches the catalogue per question and uses only the most relevant rows in each answer — how many depends on your plan, from 5 passages on the Free plan up to 20 on Business. That is why a higher plan genuinely answers catalogue questions better: it can hold more of the catalogue in view at once.

If your catalogue runs to thousands of items, split it into several AnroAgents documents by category and upload each separately. Retrieval then has a cleaner signal than one file mixing furniture with electronics.

If your catalogue changes daily, do not maintain it by hand in AnroAgents. Add the shop's website as a source on Business, which reads it again every day, use the Shopify app, which synchronises products automatically, or upload a fresh export on a schedule and delete the previous file each time.

Questions a catalogue-backed agent answers well

An AnroAgents agent with a catalogue answers descriptive questions well: what a product does, what is included, how two items differ, what fits a described need, what the price and delivery terms are.

When a visitor asks for an item with several features at once — a green velvet corner sofa, a black leather armchair — the AnroAgents agent checks each product it found against those features and offers only the products that have all of them, so it does not call a plain green sofa velvet. If none has all of them, the agent says so and explains what each close match lacks. The check reads the product's name and its labelled lines (Colour: …, Material: …, a features column), not the free-text description, so put every attribute a customer might ask for in a labelled field.

An AnroAgents agent with a catalogue answers exhaustive queries badly, and that is a property of search rather than a defect to be configured away. "Show me everything under €50" cannot be answered reliably from retrieved passages, because the agent sees the passages that matched the question, not the whole table. If customers need that, give them a filtered page on your site and let the agent link to it.

Property, vehicle and service listings

The same approach in AnroAgents works for real-estate listings, vehicles, courses and appointment types: one listing per row, every row carrying its own location, price and key attributes, and a URL column so the agent can send the visitor to the full page.

For listings that expire — sold properties, filled courses — delete the old file and upload the current export rather than adding a new one alongside it. An AnroAgents agent that still has last month's listing indexed will happily offer a flat that is gone.