ST brings coding agents to STM32 firmware ST brings coding agents to STM32 firmware

The chip vendor supplies the context

The framework helps agents choose between ST reference examples and STM32CubeMX code generation. It also reviews generated code for API use and hardware interactions. The important division is between the assistant proposing a program and the manufacturer supplying information about the chip. That gives an embedded project a more specific starting point than a general coding conversation. In its release announcement, ST described agents using its context to produce working code in a test where unassisted agents failed.

A board is part of the workflow

ST's documentation demonstrates the intended workflow with a temperature logger on an STM32C5 Nucleo board. The example combines sensor sampling, data transfer, a blinking indicator and a button that pauses output. It also asks the agent to account for documented silicon errata and demonstrate the result on the connected hardware. This makes the deliverable observable outside the editor. The proposed program has several behaviors that can be tested separately, rather than treating successful compilation as the whole result. Each project records the versions and services deployed into it.

The host determines the available tools

The installer accepts six combinations across GitHub Copilot, Claude Code and Cursor, including editor-based and terminal workflows. It writes the configuration files each agent expects. The build and DAP debugging contexts are delivered specifically through STM32CubeIDE for Visual Studio Code. That dependency makes the editor choice operational: it determines which parts of the documented workflow are present. Projects have one selected target. Moving an existing project to a different editor-and-agent pairing involves removing its deployed contexts and deploying them again for the new target.

Installation includes a licensing decision

The installation guide offers an agent-led route and a manual route. Its setup command installs compatible contexts into a chosen project, while a dry run previews the installation. ST documents separate agreements for the command-line tool, context bundles and dependent tools. Running setup, or sending the installation prompt, constitutes acceptance of the relevant agreements. The agent-led path therefore asks for the target, project folder, profile and consent before setup runs. The toolkit supplies context and tools; the chosen coding agent remains responsible for the conversation and approved actions.