Edge Agent Hardware Support Matrix: Pi, Jetson, STM32MP25, ctrlX CORE
The open-source edge-agents runtime is verified on Raspberry Pi 5, NVIDIA Jetson Orin Nano, x86 NUC, STM32MP25 and Bosch Rexroth ctrlX CORE, plus macOS for development. All verified targets run Linux. Bare-metal microcontroller execution and native OPC-UA / Modbus transports are on the roadmap, not shipping today — this page states that honestly.
This is a reference page: what runs where, with what memory, and at what status. For the install walkthrough see the Edge Agent Quickstart; for what the runtime is, see Open-Source Edge Agent Runtime.
Supported Hardware
| Device | Class | RAM | Status | Notes |
|---|---|---|---|---|
| Raspberry Pi 5 | ARM SBC | 8 GB | Supported | Reference target for the quickstart; arm64 engine build |
| NVIDIA Jetson Orin Nano | ARM + GPU | 8 GB | Supported | GPU available for local LLM acceleration |
| x86 NUC | x86 mini-PC | 16 GB | Supported | Most headroom for larger local models |
| STM32MP25 | Linux MPU | — | Supported | Linux-class microprocessor (MPU), not a bare-metal MCU |
| Bosch Rexroth ctrlX CORE | Industrial controller | — | Supported | Runs on the ctrlX automation platform |
| macOS (arm64 / amd64) | Development host | — | Supported (development) | For building and testing workflows, not a deployment target |
The engine is a ~30 MB Go service; the constraint on smaller boards is the local LLM, not the runtime itself. Use a CPU-only board (Pi 5, STM32MP25) with a small quantized model, or a Jetson for GPU-accelerated inference. The model-per-board guidance lives in Local LLMs for Edge Devices.
Roadmap (Not Available Today)
These are explicitly roadmap — do not plan a production deployment around them yet:
| Capability | Status | Notes |
|---|---|---|
| Bare-metal MCU (no Linux) | Roadmap | All current targets run Linux; true bare-metal MCU execution is planned, not shipping |
| OPC-UA transport | Roadmap | MQTT is first-class today; bridge OPC-UA into MQTT in the meantime |
| Modbus transport | Roadmap | Same as OPC-UA — bridge via a gateway today |
If you need OPC-UA or Modbus right now, bridge them into MQTT (for example with Node-RED — see edge-agents vs Node-RED), which the runtime supports natively.
What “Supported” Means Here
A device listed as Supported is one the runtime has been verified to run on. It does not imply a managed support contract. The engine builds as an arm64 (or x86) container and runs wherever Docker and a Linux kernel are available on these classes of hardware. The quickstart shows the exact arm64 build for the Pi; swapping the --platform flag retargets the build for other boards.
Choosing a Board
- Just trying it out / CPU-only inference: Raspberry Pi 5 (8 GB).
- You need GPU-accelerated local LLMs: Jetson Orin Nano.
- Largest local models / most RAM headroom: x86 NUC (16 GB).
- Industrial / automation context: STM32MP25 or ctrlX CORE.
- Building and testing workflows before deploying: macOS development host.
Where the Runtime Comes From
edge-agents is ForestHub’s open-source edge runtime. The source is on GitHub. For a production rollout across a fleet of these devices, book an architecture call with the ForestHub.ai team.
Related Pages
- Open-Source Edge Agent Runtime — what edge-agents is
- Edge Agent Quickstart — run it on a Pi
- Local LLMs for Edge Devices — which model fits which board
- edge-agents vs Node-RED — bridging OPC-UA/Modbus today
FAQ
Does edge-agents run on STM32MP25? Yes. The STM32MP25 is a Linux-class microprocessor (MPU), and the runtime is verified on it. Note this is different from a bare-metal STM32 microcontroller, which is on the roadmap.
Does it run on Bosch Rexroth ctrlX CORE? Yes — ctrlX CORE is a verified target.
Can I run it on a bare-metal microcontroller with no Linux? Not today. Every verified target runs Linux. Bare-metal MCU execution is on the roadmap.
Does it support OPC-UA or Modbus? Not natively yet — both are on the roadmap. MQTT is the first-class transport; bridge OPC-UA / Modbus into MQTT in the meantime.
How much RAM do I need? The runtime itself is light (~30 MB). RAM is driven by the local model you run alongside it. See Local LLMs for Edge Devices for sizing.