A private AI shopping assistant runs on your phone and helps you decide without telling a store, an ad network or a chatbot company what you’re buying. It can read a product label or spec sheet from a photo, compare two options side by side, explain jargon like mAh or “non-comedogenic”, and draft a return request. What it can’t do offline is check today’s price, stock or reviews. A local model has no web access, and its knowledge stops at its training date.
What do shopping apps’ AI features do with your data? #
Retail apps use AI in a few familiar ways: recommendations based on what you browse and buy, visual search where you photograph an item to find similar ones, virtual try-on through the camera, and chat assistants that answer product questions. Some of the camera processing runs on your phone. The recommendation and assistant parts almost always run on the retailer’s servers, tied to your account.
That’s the trade you make. The store learns your sizes, your budget, what you looked at and didn’t buy, and sometimes what you asked its chatbot. That data shapes the prices and promotions you see and the ads that follow you around. For everyday purchases most people shrug. For health products, baby items, gifts or anything you’d rather keep to yourself, it’s worth knowing there’s another way to get the thinking help.
What can a private AI help with while you shop? #
| Task | Private AI on your phone | Store or cloud assistant |
|---|---|---|
| Read and explain a label or spec sheet from a photo | Yes | Yes |
| Compare two products you give it | Yes | Yes |
| Explain terms (OLED vs LCD, retinol vs retinal) | Yes | Yes |
| Draft a return, warranty or complaint message | Yes | Rarely |
| Current price, stock, delivery date | No | Yes |
| Reviews and ratings | No | Yes |
| Knows products released after its training | No | Yes |
| Keeps your questions away from the retailer | Yes | No |
The honest pattern is to use both. Get prices and reviews from the store or a comparison site, then do the weighing, the “which of these actually fits my situation” part, in a private chat.
How do you compare two products with a photo? #
These steps use Personal LLM, which runs open vision models on your phone with no account.
- Install a model with image support. Qwen 3.5 4B plus its image-support file is a good start. Every model in the catalog can read photos once its vision file (195 MB to 1 GB) is added.
- Photograph the spec panel or label of the first product, straight on and in good light. Ask the model to list the specs it can read.
- Do the same for the second product. You can also paste specs copied from a product page.
- Ask for a comparison that fits you. “I mostly use my headphones on flights and calls. Which of these two is better for me, and what am I giving up?”
- Ask for a table if you’re weighing more than a few factors. The app renders Markdown tables properly.
For anything with numbers, ask it to quote what it read before comparing. Small models occasionally misread a digit, and it’s easier to catch at the quoting step. Our guide to asking AI about photos offline has more tips on getting accurate readings.
Can AI read ingredient and nutrition labels? #
Yes, and this is one of the more useful offline uses. Point the camera at an ingredient list and ask what each unfamiliar ingredient is, which ones are sugars under another name, or how two cereals compare per serving.
Two cautions. First, if you have an allergy or a medical diet, read the label yourself every time. A model can miss an ingredient or misread a tiny line, and “may contain” warnings matter. Second, ask about what’s written, not about health claims. “What does this label say about sodium per serving?” is reliable. “Is this healthy?” invites a vague answer.
Why doesn’t an offline AI know prices or new products? #
A language model learns from text collected up to a cutoff date, and a model running offline can’t browse. So it doesn’t know this week’s price, whether an item is in stock, or that a newer version came out last month. Ask anyway and it may guess, which is worse than saying it doesn’t know. We explain this in detail in does offline AI know current events.
The workaround is simple: bring the current facts to the model. Paste the two prices, the delivery dates and the return policies, and let it do the arithmetic and the reasoning.
How can a private AI help you avoid bad deals? #
- Unit prices. Paste two pack sizes and prices and ask for the cost per ounce or per item. Models are good at this; double-check the math on big purchases.
- Warranty and return terms. Photograph or paste the policy and ask what’s excluded and how long you have. It will tell you what the text says, which is often more than you’d read yourself.
- “Deal” texts and emails. A message promising a huge discount, a prize or an urgent delivery problem is a common scam pattern. Paste it into a private chat and ask whether it has scam markers. Our guide to checking whether a text is a scam walks through it.
- Writing the complaint. When something arrives broken, ask for a short, firm message to the seller with the order number, what happened and what you want. Edit and send it yourself.
Why does shopping privacy matter at all? #
What you buy says a lot: health conditions, pregnancy, finances, religion, relationships. Questions reveal even more than purchases, because you ask about things you’re only considering. When those questions go to a store’s chatbot or a cloud AI, they’re stored under someone else’s policy.
With an on-device model, the chat is computed and stored on your phone. Personal LLM has no account and no server; the only network use is the one-time model download from Hugging Face and ads in the free version when you’re online. Delete a chat and it’s gone.
Which model should you use for shopping help? #
For reading labels and comparing two items, Qwen 3.5 4B or Gemma 4 E2B is plenty and runs on phones with 3 GB of RAM or more. For dense spec sheets, long warranty documents or screenshots of product pages, GLM 4.6V Flash (needs 8 GB of RAM) is built for documents and UI screenshots. You can switch models mid-chat from the chat header, so start small and move up only when an answer looks shaky.
Frequently asked questions #
Can AI compare products for me? #
Yes. Give it the specs, prices and what you’ll use the product for, and it will lay out the trade-offs and suggest which fits. It works best when you supply the current details, because an offline model doesn’t know today’s prices or the newest models.
Does an offline AI know current prices? #
No. A model running on your phone has no internet access and only knows what was in its training data. Paste in the prices you see and it can compare them, work out unit costs and flag anything that looks off.
Is it safe to ask AI about what I’m buying? #
With a store’s assistant or a cloud chatbot, your questions are sent to and stored by that company. With an on-device model, they stay on your phone. For sensitive purchases like health products or gifts, a local model keeps the question private.
Can AI read a nutrition label from a photo? #
Yes, a vision model can read the panel and explain it, including comparing two products per serving. Always read allergy information yourself, since a small model can misread fine print.