Telegram bot builds a VkusVill grocery cart from a single phrase

One of those projects that starts with everyday frustration and ends with a working product. A Russian developer published on Habr a story about creating a Telegram bot that accepts an ordinary text phrase—for example, 'put together breakfast for two'—and independently assembles a shopping cart in VkusVill with a ready-made checkout link.

The starting point of the project is crystal clear to anyone who has ever ordered groceries online. The author describes a typical scenario: open the catalog, select milk from among a dozen and a half similar cards, then bread, then cheese, then something for tea. By the fifth item, the sense of time savings completely evaporates.

According to data from the Platforma analytics platform, Russians spend between 19 and 49 minutes per month choosing products in mobile apps, while Muscovites spend two and a half times more on this. Meanwhile, two-thirds of buyers cite time savings as the main reason for switching to online grocery shopping. The paradox is obvious: a tool created for convenience becomes a time drain itself.

Technically, the bot is built as a combination of several components. On the user side—a standard Telegram interface, no additional applications or registration. The user's message is sent to a language model deployed via Yandex Cloud AI Studio.

The key technology here is function calling, a mechanism where the language model doesn't just generate text but determines which external functions need to be called to execute the task. In this case, the model 'understands' that a breakfast request means a specific set of product categories and calls the search functions of the VkusVill catalog through its public API. Search results are aggregated into a shopping cart, and the user receives a link where they only need to confirm the order.

The author emphasizes that the project is not connected to either VkusVill or Yandex—it is completely independent development using the open interfaces of both services. The code is published on GitHub under the Apache 2.0 license, making it available for study, forks, and commercial use. For developers interested in function calling and Model Context Protocol, the project has practical value as a reference implementation—not an abstract example from documentation, but a working product with a real user scenario.

This project is interesting not so much for its own sake, but as an illustration of a broader trend. Language models are increasingly stepping beyond chat windows and beginning to control external services—placing orders, booking tables, searching for tickets. Function calling and MCP transform LLMs from conversationalists into operators capable of taking real-world actions through APIs. Retail grocery trade is one of the most obvious candidates for such automation: purchasing patterns repeat week to week, inventory is structured, and user pain from routine catalog scrolling is well measurable.

However, the approach has limitations worth keeping in mind. Public APIs of major retailers can change structure without warning, meaning the bot risks losing functionality at any moment. Personalization is currently limited by model capabilities: the bot doesn't know your dietary restrictions, brand preferences, or budget constraints unless you explicitly state them. Finally, there's the question of trust—willingness to delegate product selection to an algorithm remains a matter of individual comfort.

Nevertheless, the direction is set. If major grocery chains integrate similar interfaces into their official apps, the familiar model of online grocery shopping—endless catalog scrolling with manual item additions—could give way to a conversational format. One independent bot, of course, won't revolutionize the industry. But it clearly demonstrates that the technological stack for this already exists, is accessible, and doesn't require a team of fifty engineers. Sometimes one tired developer searching for buckwheat is enough.