Sber Launches Free GigaPlatform AI Studio for Building AI Agents Without Code

Zero rubles to get in and zero lines of code: on those terms, Sber opened GigaPlatform AI Studio on 13 October 2026. The service is positioned as a place to test an AI hypothesis on your own tasks and build an agent without programming. The intended audience is broad, from analysts and product managers to small teams with no ML engineers.

What exactly has been launched

GigaPlatform AI Studio is a layer on top of Sber's ecosystem of AI services, and the GigaChat family of models remains at its center. The idea will be familiar from Western no-code agent builders: the user describes a task, connects data and tools, configures the logic and gets a working assistant. Instead of weeks of development, a prototype comes together in hours.

One important detail in Sber's own wording: the service exists first and foremost for testing hypotheses. It is a tool for trying things out, not a ready-made production line. The company wants people to see a result on their own case first, and only then decide whether to move the task into production infrastructure.

What the initial announcement does say:

• Access is free.
• Building agents does not require programming.
• The service is designed for testing AI on the user's real tasks.
• The launch took place on 13 October 2026.

What the announcement does not disclose: the set of available models, limits on requests and data volume, the length of the free period, the list of integrations, and the terms for processing uploaded documents. All of this determines whether AI Studio becomes a working tool or stays a demo showcase.

Why Sber is doing this

Free access serves two purposes at once. The first is obvious: lower the barrier and win users while the no-code agent market in Russia is still up for grabs. Builders from Yandex, VK and several independent players already operate there, while foreign platforms remain inconvenient for Russian companies because of payments and compliance.

The second purpose is less visible. Every tested hypothesis becomes data on which tasks businesses actually want to hand over to AI: handling customer requests, parsing documents, searching knowledge bases, preparing reports. That data helps Sber understand demand and then sell paid solutions: APIs, cloud capacity and enterprise deployments.

The free tier in products like this usually works as a funnel. The user builds a prototype, gets used to the platform's interface and logic, and then runs into limits or a need to integrate with internal systems. Moving to a paid plan becomes the natural next step. This is a normal business model, but it is also exactly how you should calculate the cost of ownership up front.

Where the pitfalls are

No-code agents handle linear scenarios well: take a request, find the answer in a database, format it using a template. They struggle more with branching logic, long chains of actions and error handling. Where an agent has to make decisions on its own, quality quickly depends on the model, the prompts and testing discipline rather than on how polished the builder is.

Data is a separate question. A bank-run platform inspires trust among corporate lawyers, but that does not remove the need to check: where uploaded files are stored, whether they are used for fine-tuning, and how access control works. For companies with trade secrets or personal data, the answer matters more than the feature set.

How to start testing

If you are considering a trial, plan it so the free period yields as much information as possible:

• Pick one narrow task with a measurable result, such as classifying incoming emails or answering questions based on internal regulations.
• Prepare 30–50 real examples and define in advance what counts as a correct answer.
• Do not upload sensitive data until the terms of its storage and use are clarified.
• Record how long the prototype takes to build and the share of correct answers, so you can compare against alternatives.
• Ask Sber about the terms after the trial stage: pricing, limits, and whether you can export the agent and its logic.

The bottom line is simple. If within a day or two you have an agent that handles at least part of your routine work with acceptable accuracy, the hypothesis is alive and can be scaled. If not, you spent time rather than budget, and that is exactly why such services are made free.