finds and fills in the parameters and images, checks the facts and writes the descriptions

It finds and verifies the product information, fills in what is missing and writes the descriptions from what it has checked. It runs on our own computers in Europe, so we know what it does with your data.

TomášMichal
We have been doing AI for e-commerce for four years. We came to the conclusion that the part that matters we have to build ourselves.
Tomáš & MichalCo-Founders

Hundreds of e-shops already run on Pobo

What Pobo LLM can do

The machine is running and this is the work we are moving onto it.

Product categorisation

Thousands of products sorted overnight

Sorting products into categories and filling in parameters across the whole catalogue. Exactly the kind of job that is unaffordable on a foreign API and routine on our own machine.

Looking up parameters and images

Facts that at least two sources agree on

Pobo LLM looks up missing parameters, images and other product details on supplier and manufacturer sites, compares the sources and fills in only what checks out.

Generating product descriptions

Rich content in the look of your shop

Descriptions written from verified information and placed into your design template, FAQ and galleries included. Thousands of products without paying a foreign API for each one.

AI listing of products

Without retyping a single row

A supplier file becomes finished product pages. Reading the file, matching the columns and filling in what is missing moves onto our own model.

Agent orchestration

One step fails, not the whole job

Instead of one big request, the work is split into steps that the system runs, checks and repeats when one of them fails. You see which step a job is on, not just a spinner.

Tracking your shop's visibility in AI

You find out whether the assistants name you

Pobo LLM asks the assistants people actually use, watches whether your shop comes up in the answers, and fills the facts that are missing into your descriptions.

Want one of these first? Tell us on a demo. Book a personal demo →

How Pobo LLM works

Three of them up close, each with a different shape of work.

Pobo LLM1Supplier fileCSV, XML, Excel, TXT2Supplier filemapping3Looking upfacts andproductinformation4Writingproductdescriptions(rich content)5Contentchecked by aperson6Supplier fileexport

AI listing of products

From a supplier file to finished product pages

The file goes round once. Pobo LLM reads it, matches the columns to your catalogue, looks up what is missing, writes the descriptions, checks the result and hands over finished product pages.

Pobo LLM1Choosingproducts todescribe2Choosing aprompt forcontentgeneration3Choosing adesigntemplate4Generatingproduct imagesand icons5Approving thedescriptions6Export to theshopCSV, Excel, XML or API

Generating product descriptions

From choosing products to exporting to your shop

You choose the products, the prompt and the design template. Pobo LLM writes the descriptions and the missing product images and icons are added alongside. You approve the result and it goes to your shop.

How description generation works →
Pobo LLM1We ask like acustomer would2We measure AIvisibility3We checkwhether youare in theanswer4We find whatis missingabout you5We write itinto yourdescriptions

Tracking your shop's visibility in AI

Measures whether the assistants name you and fills in what is missing

Pobo LLM measures your visibility in AI assistants, works the findings into the descriptions and measures again.

Sára
In Pobo we do not just generate text. Complete content comes out of it, FAQ, galleries and videos included, and all of it holds the look of your shop.
SáraUX designer

What you get when the AI runs on our machines

Three things a rented AI cannot give you, no matter how much you pay for it.

Your data stays in Europe, on hardware we own

Supplier price lists, catalogues and internal documents are processed on our own hardware. They never travel to a foreign cloud, so when a legal department asks where the data sits, there is one concrete answer.

No ceiling on volume, or on how often you run it

Our own machine does not count requests. A job that goes through your entire catalogue can run overnight, and if the result is not good enough, you run it again without paying anything extra.

Independence from foreign providers

Rate limits, price changes and models being switched off somewhere else do not concern us. What runs here keeps running.

What we avoided with AI

The model itself is the easy part. What makes the result usable is everything around it.

  1. We are not a wrapper around one model

    What if one model gets expensive or disappears?

    For every task we pick what suits it. Our own model handles the data and writes the product descriptions, the best available models take care of translations and articles. That choice is ours to make, not something a contract dictates.

  2. The AI does not get the last word

    What if the model makes up a price?

    Where a mistake costs money, a fixed rule decides, not the model's guess. A price has to be a number, an EAN has to have eight or thirteen digits. No model talks its way past that.

  3. One source is not enough for us

    How do you know the information is true?

    When we look something up about a product, we compare several independent sources and use only what at least two of them agree on. One model makes things up. Three do not make up the same thing.

How we keep your data safe

Agents bring the data in, Pobo LLM decides what gets through. Nothing lands in your shop without passing the check.

Secure layerPobo LLMReader agentCrawler agentScout agentSync agentGuard agentReading thesupplier fileFields mapped,products draftedWeb searchSupplier andmanufacturer sitesAsking theassistantsReading thecatalogueProducts, categories,brandsHandling imagesand filesUploads from themerchant

The model never touches your database

Think of Pobo LLM as a copywriter without the keys to your shop. It gets the source material, hands back finished text, and that is where its job ends. Saving the text into your shop is done by our system, and only where it belongs.

Every step stays in the log

For every generated text we keep the prompt that produced it, the model that wrote it and the moment it happened. When something comes out wrong, it can be traced back.

Vojta
A supplier file, text from the web, an image someone uploads. We treat all of it as unverified. It reaches your shop only after it has passed our checks.
VojtaAI engineer

What runs where

A plain list, so you can forward it to whoever asks.

What runs on Pobo LLM

High volume, sensitive input, and a result that goes through a check before anyone acts on it.

  • Reading supplier files and price lists
  • Sorting products and filling in parameters
  • Matching duplicate products across catalogues
  • Generating product descriptions
  • Quality checks before a description is published
  • Internal analytics over your own data

What we still run on the best available models

Text your customer reads. We are not going to make it worse just to have a nicer marketing claim.

  • Blog articles
  • Translations into other languages
  • Looking information up on the web, where the value is in the sources

For larger e-shops

The answers procurement and IT ask for, not marketing.

Where does the server physically stand?
In Czechia. Production in Prague, backup in Brno, a physical server rather than rented cloud. Exact placement and technical details come with the security questionnaire.
Can we get our data out?
Yes, over the REST API. It lists products, categories, blogs and brands in every language, and returns both the finished HTML and the underlying widget structure. Name and URL lists can also be downloaded as CSV straight from the admin.
Can we connect our own system?
Yes. Pobo has a REST API and an MCP server, so both your own system and the agent you use can reach the content. Direct access to the model itself is not something we offer yet.
Who has access to the data and is there a contract?
We send the data processing agreement on request. Thanks to our own infrastructure it names a noticeably shorter list of sub-processors than tools built on somebody else's cloud.
What happens if we leave?
The export over the API works for as long as you have an account, including on the last day. What you created in Pobo is yours to take with you.

Frequently asked questions

Where does the machine actually stand?

On our own hardware in Europe. It is not a rented slice of somebody else's cloud.

So my data never leaves Pobo?

For the tasks that run on Pobo LLM, yes, and writing product descriptions is one of them. Translations, blog articles and looking things up on the web still go through external models, and we would rather say so than promise something we cannot keep. The list above shows exactly which is which.

Will it make Pobo more expensive?

We will name the price of each feature when it is finished. What we can say now: jobs that go through an entire catalogue are the most expensive thing a foreign API charges for, and exactly what our own machine handles overnight for the price of electricity. That is why we are building on it.

When can I use it?

Now. Pobo LLM is in operation on our own hardware and runs underneath what you already use. Tell us you are interested and we will go through your catalogue together.

Do I have to set anything up?

No. It runs underneath what you already use. You will notice it as things getting faster and fewer mistakes left to fix by hand.

Let us show you AI on your own products

In thirty minutes Lenka goes through your catalogue with you and shows you what the descriptions would look like. If you would rather look around on your own, you can start right away.

  • We go through your catalogue together
  • You see which jobs run on our own hardware
  • No commitment and no data written anywhere
Lenka