Products
Store what you sell once — name, price, currency, description, notes — and every agent knows it. Sync Shopify and your catalogue stays read-only to the AI.
Cette page n'est pas encore disponible dans votre langue : voici la version anglaise.
Every price you've ever typed into a chat window died there. Type it into your Neural Core instead and it becomes a record: a real name, a real price, a real currency, a description and your private notes. All ten agents read it on every single run — not because they were reminded, but because your Neural Core is loaded into every agent's context before they say a word.
The difference shows up the moment you stop describing your product and start using it. Ask for a post about it and the image generator reaches for your actual product photos. Ask for content ideas and every one comes back tied to a real product name from your catalogue — no "new collection", no "behind the scenes".
What a product record holds
| Field | What it's for | Notes |
|---|---|---|
| Name | The identifier everything else keys off | This is the string agents match against when you mention it in a prompt |
| Price | Digits only — 49.90, not €49.90 | Kept as a real value, not prose |
| Currency | EUR, USD, GBP… | Stored beside the price so agents don't guess |
| Description | What the product is, in your words | This is what lands in captions, emails and documents |
| Notes | Internal context — margin, stock quirks, who it's for | Yours. Lock it (see below) and it never reaches generated content |
| Images | Product photos | Uploaded in the app — the extraction pass never writes or modifies them |
Prices and currencies being real fields, rather than a sentence in a paragraph, is the whole point. It's why an agent can list your catalogue back to you cleanly, and why a spreadsheet export makes sense.
Add a product
By talking. The fastest route — say it to any agent, in any language:
Add a product to my Neural Core: "Signature Roast", 18.50 EUR, single-origin Ethiopian, 250g bag. Internal note: our best margin item.
The agent writes the record and confirms in one short sentence. No command, no prefix, no /remember.
By hand.
- Open the Neural Core from your sidebar.
- Click the Products card in the hero grid.
- Add an entry and fill in name, price, currency and description.
- Upload product photos on the entry — these are what image generation will reach for later.
By import. Drop a file straight onto the Products card and it gets parsed into real product records:
- Open Products in the Neural Core.
- Drag a
.csv,.xlsx,.xls,.ods,.tsv,.pdf,.docx,.txtor.mdfile onto it. - The importer detects the separator on its own and fuzzy-matches your column headers.
- An import folder is created automatically and your products land inside it.
Your headers don't have to be in English
The importer matches product column headers across French, English, Spanish, German, Turkish, Portuguese and Italian. Export your catalogue however your tool exports it — you shouldn't have to rename columns to talk to your own memory. Files cap at 50 MB.
Let it learn while you talk
You don't have to file anything. Say "actually the Signature Roast is 19.50 now" mid-conversation and the extraction pass picks it up, decides on its own whether that's a new product or a correction to an existing one, and updates the record — in whatever language you said it in.
Two deliberate limits, so you know what to expect:
- It won't invent placeholders. Entries called "product name", "offer" or "service name" are rejected outright.
- It won't touch your images or your product URL. Both are protected fields — extraction never writes them. What you uploaded stays what you uploaded.
- It won't learn from the AI. The extractor sees your messages in full but only ever sees an 80-character topic hint of the agent's reply, and is explicitly forbidden from extracting facts the assistant stated. Your catalogue can't fill up with things the model made up.
Not everything said is captured
Extraction runs against a confidence floor and emits a limited number of items per pass. It's very good, not omniscient. If a price matters, say it plainly — or check the Products card. "It remembers everything" would be a lie, so we're not telling it.
Shopify: synced, and locked on purpose
Connect Shopify and your products, customers and orders sync into the Neural Core automatically. Synced products are marked as coming from Shopify — and every agent write tool refuses to touch them.
That refusal is the feature.
- Shopify is read-only by grant. All seven scopes Unyo requests are
read_*. There is no code path in Unyo that writes to your store. - Synced products are read-only in Unyo too.
updateanddeleteare rejected on any Shopify-sourced product, by every agent, at the tool level — not by a prompt asking nicely. - So your store stays the source of truth. Change a price in Shopify; the sync brings it over. Nothing you say to an agent can quietly desync your catalogue from your storefront.
If an agent tells you a product is locked, that's this. Edit it in Shopify.
What syncs, and how much
The sync brings over up to 100 products, 100 customers and 250 orders, and paces itself against Shopify's rate limit so it never trips your store's API budget. A catalogue bigger than that won't come across whole — add the products you actually want your agents talking about by hand.
See the Shopify integration guide to connect.
Products make images better — automatically
This is the part people don't expect. When you ask for an image and your product's name appears in the prompt, Unyo notices, finds that product in your Neural Core, and feeds its real photos to the image generator as references. You don't ask for it. There's no toggle.
Because "Signature Roast" is a real product with real photos on file, the generated image is built around your actual bag — not a stock-photo idea of a coffee bag. Any visual style preferences you've saved on that product get applied on top.
Three conditions, so you can debug it when it doesn't fire:
- The product's name must literally appear in your prompt. "my coffee" won't match; "Signature Roast" will.
- The product must have images uploaded on its Neural Core entry. No photos, no product-aware generation.
- If two product names both appear, the longest match wins.
More on generation in Images.
Products make content ideas real
Ask any agent with your social channels for ideas and you get up to five — each one tied to a specific product from your catalogue, informed by your own top posts, your own hashtags and your own peak times.
The instruction behind that is deliberately strict: every idea must name a real product, and generic filler — "new collection", "behind the scenes", "lookbook" — is banned outright. If your Products card is empty, this is the feature you're leaving on the table.
Keep the internal stuff internal
Some things about a product are for you: margin, supplier, the reason you're discontinuing it. Put them in Notes and mark the entry Private with the lock control.
A locked entry isn't hidden from your agents — it's flagged inside their prompt as do not mention in any generated content, emails, or posts. Every agent, simultaneously. They can reason with it; they can't publish it.
Ask, edit, export
Every agent can work your catalogue conversationally:
Search reaches inside product descriptions, not just names — so "washed process" finds it even if that phrase is buried in the description.
Listing caps at 30
When an agent lists your products, it returns up to 30 at a time, and an export returns up to 200. If you sell more than that, ask for what you need — by name, by search — rather than "list everything". The Products card in the app shows them all.
Deleting a product
Ask an agent to delete one and it's gone — and it stays gone. Deleting writes an invisible tombstone that gets re-injected into the extraction pass as "the user deleted this — do not re-extract". Systems that quietly re-learn what you just removed are the norm. This one is built not to.
Deletion is permanent, and product delete isn't hard-gated
There's no undo and no version history on a Neural Core product. Agents are instructed to ask for one short confirmation when your request is ambiguous — but on products specifically that's an instruction, not a technical lock. Be explicit about what you want deleted, and prefer deleting from the Products card when you're removing several.
Products cost nothing
Every Neural Core tool — reading, saving, updating, listing, searching, exporting your products — costs zero credits. All of them. You're only ever billed for the agent run itself, at real cost. Your memory is not a metered upsell.
Troubleshooting
"That product is locked / read-only." It came from Shopify. That's by design — your store is the source of truth. Edit it in Shopify and let the sync bring it across.
The generated image ignored my product photos. Check the three conditions above: the product's exact name has to be in your prompt, and the product entry has to have images uploaded on it. A product with no photos can't be used as a reference.
My import created products with mangled names. The importer combines separate first-name/last-name style columns and fuzzy-matches headers, but a genuinely unusual export can confuse it. Open the import folder it created and fix the entries directly — they're real, editable records.
An agent said it updated the price but the card looks the same. Refresh the Neural Core page. If it's still stale, ask the agent to read the product back to you — that returns the stored record, not what it remembers saying.
An agent won't list all my products. That's the 30-item cap. Ask by name or by search term instead.
Do I have to add products for Unyo to be useful?
No — but the difference is large. Without products, image generation works from your prompt alone and content ideas have nothing real to anchor to. With products, the image generator uses your actual photos and every idea names something you actually sell. It's the highest-leverage ten minutes in the whole app.
Can an agent change my Shopify catalogue?
No, in two independent ways. Unyo only ever asks Shopify for read permissions — all seven scopes it requests are read-only, so writing isn't something the connection is allowed to do. And separately, inside Unyo, every agent write tool refuses any product that came from Shopify. Your storefront can't be edited from a chat window.
Which agents can see my products?
All ten. Your Neural Core isn't mounted per-agent or gated by a setting — it's loaded into every agent's context on every run, and the code that builds that context doesn't even take the agent's identity into account. Tell one agent your price; they all know it.
How do I stop an agent mentioning a product's internal notes?
Mark the entry Private with the lock control. The entry is then flagged inside every agent's prompt as not to be mentioned in any generated content, email or post. It's enforced where it matters — in the prompt — not just greyed out in the interface.
Can I get my product data back out?
Yes. Ask any agent to export your products as CSV or JSON (up to 200 entries) and to save the result to Drive or Notion, or download it yourself from the Neural Core. It's your catalogue.