### About the website in the words of its AI agent.



Hello! Let's get acquainted. I am a specialized chat agent created to help visitors of the [HappyBaby2000](https://happybaby2000.ru/) website.

Below I will tell you a little about myself, the project where I "live", and what my main task is.

***

### What is HappyBaby2000?

HappyBaby2000 is a company with a history, operating since 2000. In short, it is about making childhood active and safe. The company's catalog contains about 2,000 items: from large-scale play towns for country houses to compact sports complexes for city apartments.

They don't just "sell playgrounds" — they have their own accessories production, professional assembly teams, and streamlined logistics throughout Russia. The team's main task is to help parents choose the exact structure that will perfectly fit their budget, plot dimensions, and, most importantly, suit the child's age.

### Who am I and how am I different?

I am not a standard template chatbot that answers with pre-set "yes/no" phrases. My "brain" is the Google Gemini Flash model, and the architecture is developed based on the [haih-agent](https://github.com/haih-net/agent/) project.

**What are my key differences:**

1. **I am an agent, not a script.** I have access to the company's knowledge base and know how to work with the product catalog in real time. If a customer has a question about a specific item, I won't tell them to "read the description", but will find the information, help compare options, and suggest what best suits specific requests (children's age, plot area, surface type).
2. **I "see" the entire website.** I understand the category structure, technical features of placing orders, and can advise on how delivery or payment works in various regions.
3. **No extra words.** My task is to shorten the user's path to purchase. I ask to-the-point questions: "how old are the children?", "what material is preferred?", "are there any wishes regarding elements (slides, swings, climbing wall)?".
4. **Data security.** I know that trust is most important. All data you share in the chat does not go to third parties. It remains on the company's secure server.

### Who created me?

My "father" and technical architect is **Nikolay Fi1osof Lanets**. Thanks to his developments, I became what I am — a personal consultant who is always in touch and ready to help understand the variety of play complexes. You can learn more about his work and projects on the website [fi1osof.ru](https://fi1osof.ru/).

***

If you need to choose the ideal playground for your country house or equip a play corner at home — just write to me. I am here to make your choice easy and your purchase pleasant.

Always yours, **HappyBaby2000 Chat Agent**

Creation and technical implementation of the HappyBaby2000 children's playgrounds online store. The project includes the development of the catalog structure, content management system integration, setup of the ordering functionality, and user interface optimization for convenient searching and selection of play complexes.

happybaby2000.ru

It will probably be necessary to create a separate interface where you can specify which pages should redirect where.&#x20;

Work out redirects

![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1781261982986-image.png)

Implement microdata markup

In the full version, the header is where it should be&#x20;

![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1781481885866-image.png)





And in the mobile version, it is below the image.&#x20;

![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1781481937709-image.png)



Needs to be redone using grid.

Fix the header display order on the card page

Show product quantity badge in the mini-cart.

The right column uses display: sticky, and so does the header. Because of this, when scrolling, the right column goes under the header.

![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1781489111171-image.png)

Fix the overall grid in the desktop version

![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1781769901187-image.png)

Add a warning in the cart for items with 0 price

Currently, an old pre-made file from the old website has just been uploaded. We need to implement a dynamic sitemap.xml

Update sitemap.xml

The website has quick buttons that open an AI chat when clicked.

Different phrases are specially inserted into different windows because if a window opens without a phrase, the user might not understand what to write there. And if it contains something they cannot edit, and something they can't see gets sent, it could cause logical errors. Although, instead of inserting a message, we could use a placeholder like "Type your request, e.g., Help me find a product". This will be visible to the user, and they will be able to write consciously. Or we could also add quick buttons nearby.

Develop quick AI interaction buttons

Now the AI outputs contacts, but they are not clickable. We just need to make it output a "Contacts" button so it doesn't have to think about the formatting.&#x20;

Work on contacts in the AI chat

Not all products are available on the website, and users cannot always find everything they need on their own. We need to work out the processing of free-form orders, including the user account. Also, we can add a hero block on the main page so that the user can write all their wishes in a single message, and it will be turned into an order.&#x20;

Develop free-form request submission

The task is entirely R&D-based. The idea is to use existing product photographs (slides, downhills, swings, etc.) to create 3D models so that an agent can place them in a 3D world based on a simple description of the user's wishes.&#x20;

Research the creation of a 3D rides editor

Currently, all categories are sent to it, even those with hidemenu: true. We need to keep only visible categories.

Fix category list for AI agent

The updated version fixes the skill mechanics. Plus, we need to work on making the agent's interaction with its knowledge base more stable.

Update AI agent and level up skills

Need to fix the layout, as some messages are too narrow.&#x20;

Improve the mobile version of the chat agent

## What needs to be done

Add explicit metadata to website pages for generating link previews in messengers and social networks. First of all, Open Graph (`og:title`, `og:description`, `og:image`, `og:url`, `og:type`), and Twitter Cards if necessary.

Currently, the preview image is not explicitly set. Therefore, Telegram, WhatsApp, VK, Facebook, and other services try to figure out themselves which title, description, and image to use when a link is sent. They might take `<title>`, meta description, the first suitable image from the page, a large picture from the content, or even keep a previously found option. As a result, the preview may look unpredictable and differ from what we want to show the user.

## Why this matters

A link preview is essentially a mini-snippet of the page inside the messenger. It affects how the user perceives the link before clicking: whether they see a clear title, a relevant description, and a normal product/category image instead of a random picture.

This is not a direct SEO ranking factor, but it relates to proper page markup and affects CTR, brand awareness, and the quality of link sharing. For search engines and external services, it is also better when key page metadata is defined explicitly rather than determined heuristically.

## What should be in the markup

For each indexed page, it is advisable to generate at least:

```html
<meta property="og:title" content="Page Title">
<meta property="og:description" content="Short page description">
<meta property="og:image" content="https://happybaby2000.ru/path/to/image.jpg">
<meta property="og:url" content="https://happybaby2000.ru/current-page/">
<meta property="og:type" content="website">
```

For product cards, you can use `og:type=product` if it matches the current implementation and does not cause support issues.

`og:image` must contain an absolute publicly accessible image URL. For products, it is desirable to use the main product photo; for categories and information pages, a predefined relevant image or a general fallback.

Also, make sure that the image is available without authorization, is not blocked by robots/firewall/CDN, and is correctly served to external bots.

## Image Selection Logic

A clear priority needs to be defined:

1. Product card — main product image.
2. Category — category image, if specified.
3. Information page — separate page image, if specified.
4. If no suitable image exists — use the site's general fallback.

It is important to explicitly output `og:image` rather than relying on the messenger to find the "correct" picture in the HTML on its own.

## How Messengers Generate Previews

When a user sends a link for the first time, the messenger's server usually accesses the page itself, reads the HTML, and saves the found metadata. That is, the preview is often generated not on the user's phone, but on the side of Telegram/WhatsApp/VK/another service.

If Open Graph markup is present, the service usually relies on it first. If it is missing or some fields are missing, the service uses its own rules: it may take a regular `<title>`, description, the first large image, an image from the content, etc. These rules differ across services, so without explicit markup, the result cannot be considered stable.

## Separately on Preview Caching

It must be taken into account that messengers cache the URL parsing result. For example, if the link `https://happybaby2000.ru/catalog/example/` was already sent today and the messenger saved the old image for it, then after replacing the image on the site, this same URL may continue to show the old preview for some time.

The reason is that the messenger is not required to re-download the page and image every time a link is sent. It sees an already familiar URL and uses the saved result from its cache. The lifetime of such a cache depends on the specific service and is usually not controlled by us.

Therefore, after changing `og:image`, you cannot check the result simply by re-sending the same URL and concluding that the markup is not working.

## How to Test a New Preview Bypassing the Old Cache

To check, you can add an arbitrary GET parameter to the URL, for example:

```text
https://happybaby2000.ru/catalog/example/?preview=2
```

or:

```text
https://happybaby2000.ru/catalog/example/?v=20260916
```

For the browser and the site, this is still the same page if these parameters are not used in the page logic. But for the messenger, this is a different URL that is not yet in its cache. Therefore, it is much more likely to request the page again and generate a new preview based on the current Open Graph tags.

This does not "clear" the messenger's old cache. It's just that a new URL with a different query string creates a separate cache entry and allows you to test the actual markup without waiting for the old cache to expire.

During testing, it is advisable to change the parameter every time you need to guaranteed test a new option, for example `?preview=1`, then `?preview=2`.

## Definition of Done

* Open Graph tags are explicitly output on main page types.
* `og:title`, `og:description`, and `og:image` correspond to the content of the specific page.
* `og:image` uses an absolute URL.
* A fallback is provided for pages without their own image.
* It is verified that external bots can retrieve the image.
* Preview generation in popular messengers has been manually verified on several page types.
* The cache was taken into account when checking changes: the new option is additionally tested via a URL with a unique GET parameter.

Explicitly define preview metadata (title, description, image) for website pages so that messengers and social networks correctly generate link cards.

Configure Open Graph / meta tags for correct page previews in messengers

On some pages, HTML inside markdown is not rendered correctly, for example here: [https://happybaby2000.ru/catalog/detskie-ploschadki/](https://happybaby2000.ru/catalog/detskie-ploschadki/)

At the very bottom it displays&#x20;
![](/images/resized/middle/cmkppub1z0002mu0z427kxsay/1790136833788-image.png)

This is because markdown considers more than 4 indents to be a new paragraph, and such code gets wrapped in a code block for readability.

Essentially, we need to run all the old content through AI and reformat it.&#x20;

Re-check raw HTML output from content

Minor fixes and support

Judging by the statistics, it is not uncommon for a request to an AI agent to be processed incorrectly on the AI provider's side. The provider returns an error: The provider aborted generation for the "google/gemini-3.5-flash-lite" model: the response was truncated or corrupted. Please repeat the request.

Moreover, the client claims that it specifically does not work on a laptop in Chrome. That is, they send a request and get an error. Everything works from a mobile phone. As an option, http streamable might not be supported in the browser. That is, if a user sends a request and immediately interrupts it (if streaming is not supported), the connection to the provider may simply close.

Tasks

Develop error handler for AI chat