Sber Develops GigaChain Without Forking LangChain, Betting on a Compatibility Package

Sber shares how it's building an open source ecosystem around GigaChat and why it opted against a LangChain fork in favor of a separate compatibility package. This shift has simplified AI agent development within the company and made the stack more comprehensible for external teams.

Why the approach changed

GigaChain emerged when Sber needed a unified tool for developing with GigaChat — both for internal teams and external developers. Rather than building everything from scratch, the team decided to leverage an existing open source foundation. LangChain was chosen as the base: it was the most prominent framework for agent systems, capable of working with different models without locking developers into a single vendor. For a large company, this choice meant less risk, faster implementation, and a more familiar talent market.

Initially, the team did pursue a fork approach. The reason was practical: the era of prompt engineering was only then gaining momentum, and LangChain had numerous prompts hardcoded in English. For GigaChat, which was better adapted to the Russian language, this created unnecessary constraints. An additional challenge was that implementing full multilinguality and vendor integration within the main project was difficult to push through quickly.

But this approach soon became too costly to maintain.

"We were forced to pull in 100–200 changes that appeared in the main

project literally every week."

When LangChain began changing rapidly and transitioning to a modular architecture, Sber abandoned the fork and built its own compatibility package. Now the recommended scenario works like this: a developer installs clean LangChain and adds the GigaChat integration package. This approach proved more sustainable: even after the package dropped from LangChain's official documentation, users experienced almost no change. Effectively, the team maintained compatibility with the market standard without getting stuck in endless synchronization of external code.

What Sber is building

Around GigaChat, the team is developing not just one SDK, but an entire open ecosystem. Its goal is not only to provide API access to the model, but also to shorten the path to a working AI agent. According to the team, a significant share of Sber's internal agent projects already rely on this stack, and external developers are using it increasingly. A separate signal of demand is the Python library for working with the GigaChat API: according to ClickPy data, it ranks in the top 1.5% most downloaded packages on PyPI by monthly downloads.

• Python library for GigaChat API integration
• LangChain compatibility package instead of its own fork
• open autonomous agent GigaAgent
• documentation, integration examples and practical guides
• webinars, articles, and conference demonstrations

GigaAgent plays a special role here. The team didn't build it as a lab demo project, but rather refined it based on feedback from engineers and business teams. Throughout 2025, at dozens of conferences, developers demonstrated the agent live and gathered input: where REPL and tools were needed, where local execution was critical, and where modularity mattered most. Business, in turn, pushed toward practical scenarios — data analysis, presentation preparation, and work with corporate data sources.

What open source provides

For Sber, the open approach solves several problems at once. First, it's simpler to hire specialists: if the stack is built around LangChain and compatible libraries, the company doesn't need to find people for a completely unique internal platform. Second, the barrier to migrating existing agent solutions to GigaChat is lowered. The idea is straightforward: if a developer has already built an agent on a popular framework, it should be easier to adapt it to the Russian model without a complete rewrite.

There's a third effect—community help. The team gave an example with LlamaIndex: initially, Sber didn't support this framework due to limited resources, but later an external developer sent a pull request with GigaChat support based on an already existing library. As a result, the ecosystem gained integration with another popular stack at almost no internal cost.

The licensing policy remains as simple as possible: for its projects, the team typically chooses MIT and carefully monitors licenses and the behavior of external maintainers.

For Sber, communication around open source is not a side activity but part of the product. The team responds on GitHub, maintains documentation, publishes articles, runs Telegram channels, holds webinars, and appears at specialized conferences. In 2025 alone, it has made over 15 external presentations.

In this model, content is needed not for PR purposes, but to reduce the time between release, feedback, and the next product iteration.

What this means

The GigaChain story shows that for corporate AI stacks today, what matters more than a fork at any cost is compatibility with mainstream open source tools in the market. If Sber continues on this course, GigaChat can significantly strengthen its position in Russian-language agent system development and in corporate scenarios, where implementation speed and access to a familiar stack often matter more than technological exotica.