Why is STM32Cube for Agentic AI released as a beta?¶

STM32Cube for Agentic AI is delivered as bundles by STM32CubeCLI, which has run inside STM32CubeIDE for Visual Studio Code since April 2025 and inside STM32CubeMX2 since March 2026. The tools that the contexts drive, such as STM32CubeMX2 and STM32CubeProgrammer, are released on their own.

What is still settling is the agentic layer: the list of contexts and services, the names of the skills and of the cube agentic-ai commands, and the guidance that the contexts give to the AI coding agent. New contexts and fixes ship as feedback comes in.

What this means for you¶

The tools that the contexts drive keep their own release status.
Context, service and command names can change before general availability. The release notes list each change.
A project keeps the context versions it was deployed with. Updating the machine does not change a project until you deploy again.
The AI coding agent and its model produce the answers. The toolkit does not itself perform AI inference or make autonomous decisions.
The toolkit is for evaluation. No target is production-validated.
Sidekick KB and Errata Hunter use proven data, but their MCP services remain beta.

Known limitations¶

For the current connector inventory and version-specific limitation table, see Supported contexts and versions and Known limitations.

Validation scope¶

Six editor-and-agent combinations are accepted, but integration testing covers only the pairings listed in Known limitations. The other combinations are tested for file placement. The current restrictions and workarounds are listed in Known limitations and Supported IDEs and AI coding agents.

Answer quality¶

The toolkit provides STM32Cube-specific context and tools, but response quality depends on the selected AI agent and its underlying model. Results can vary with the agent, model, and supplied project context. Review generated changes and verify technical results in the target environment.