STM32Cube for Agentic AI
STM32Cube for Agentic AI ¶
Beta release — why?
Ask an AI coding agent to turn an application idea into STM32 firmware: configure peripherals, check documentation and silicon errata, build and program the firmware, and verify its behavior on a connected board. STM32Cube for Agentic AI provides the technical contexts that connect the agent to the STM32Cube ecosystem.
Quick start ¶
Check Prerequisites and choose an editor and AI coding agent in Supported IDEs and AI coding agents.
Follow the installation guide to install and deploy contexts into a project folder.
Open that folder with the selected agent and ask:
List the STM32 services deployed in this project.Use one service for a read-only question and ask the agent to cite its source or evidence. See Try it out for a first-use example.
Explore the toolkit ¶
Choose the contexts needed for a project. The six supported editor-and-agent combinations describe host setups, not the number of contexts. See STM32Cube Agentic Contexts for the context inventory.
Choose between deploying all compatible contexts or selecting them individually.
Follow the quick-start guide for an agent-led or manual installation.
Browse each context, its scope, and its detailed user guide.
Check supported targets, prerequisites, current versions, and limitations.
Find answers about AI assistants, project setup, and common issues.
Understand the beta scope and how the toolkit evolves.
Try it out ¶
Copy and paste this prompt into your favorite coding assistant, augmented with the STM32Cube for Agentic AI contexts:
I have an STM32C5 Nucleo plugged in. Build me a small temperature logger on it.
Sample the MCU's internal temperature sensor with ADC1 through DMA and print
one line per second over the virtual COM port at 115200, formatted exactly as
"temp=NN.N C". Use the factory calibration values so the reading is in real
degrees, not raw counts.
Blink the user LED at 2 Hz so I can see the system is alive. When I press the
user button, pause the logging and print "logging paused". Press it again to
resume and print "logging resumed".
Before you settle the design, check whether any documented silicon errata
affect the way you are using the ADC or DMA on this device. If any do,
implement the recommended workaround and tell me the errata IDs you applied.
Leave me a short handoff note describing what was configured and why.
Get it running on the board and show me it works.
Note
The toolkit is in beta: see Why the toolkit is in beta. We use feedback to improve the contexts and publish updates continuously. Context bundles can be updated independently. The installer is the central place to discover and deploy them.
Project-based deployment ¶
Deploy contexts into a project or workspace, not into the user home folder. This lets each project use its own à-la-carte context selection. Only the context bundles are installed at machine scope. Each project records its deployed versions and services.
Bundles and scope ¶
STM32Cube Agentic Contexts are distributed as independently versioned bundles through STM32CubeCLI. A context can use packaged data, project files, or a remote MCP service. The selected AI agent remains responsible for the conversation and approved actions.
For bundle concepts, see the STM32CubeCLI documentation. For legal information and installation, see the installation guide.