Yandex B2B Tech launched the SourceCraft CLI console application on April 22, 2026, bringing command-line control to its cloud platform and placing an AI agent directly in the developer's terminal. The agent writes code, runs tests, and opens pull requests. Tasks that once demanded hours now fold into a single typed instruction.
What the tool does: The SourceCraft CLI accepts natural-language prompts and translates them into working code. Developers describe a task, refactor a module, add validation logic, wire up an endpoint, and the agent executes it autonomously. It writes the implementation, validates the output with automated tests, and submits a pull request for review. The workflow collapses manual steps into a single command.
According to the April 22, 2026 announcement, a Yandex B2B Tech press office statement explained,
"In the console app, the developer states a task in natural language, after which the agent executes it autonomously in the background, writing code, validating it with tests, and creating a pull request."
Architecture and model support: Yandex AI Studio powers the agent, offering access to proprietary language models alongside open-source options including DeepSeek, Qwen, and GPT-oss. Code and work data remain inside the platform, a design choice that matters for U.S. teams navigating compliance requirements and internal security policies. The architecture keeps the development environment contained, reducing the surface area for data leakage.
What comes next: Dmitri Ivanov, head of the SourceCraft platform, outlined the next development stage. The system will integrate a team's accumulated knowledge, past technical decisions, internal standards, architecture patterns, and apply them to new tasks. The agent will learn not only from external models but from the organization's own history, accelerating work that depends on institutional memory.
The evolution from general-purpose code generation to context-aware execution represents a meaningful shift in how development tools understand the difference between writing code and writing the right code. The CLI is a step toward tools that know what a team has already decided, promising to reduce onboarding friction and enforce consistency across distributed teams working in complex codebases.
