diff --git a/examples/arduino/.gitignore b/examples/arduino/.gitignore
new file mode 100644
index 00000000000..453c723c671
--- /dev/null
+++ b/examples/arduino/.gitignore
@@ -0,0 +1,2 @@
+# Generated by build_arduino_library.sh — do not check in
+arduino_lib/
diff --git a/examples/arduino/ExecuTorchArduino.h b/examples/arduino/ExecuTorchArduino.h
new file mode 100644
index 00000000000..dc7e87706fd
--- /dev/null
+++ b/examples/arduino/ExecuTorchArduino.h
@@ -0,0 +1,24 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+#pragma once
+
+// Arduino's custom <new> header omits <exception>, which breaks
+// std::bad_variant_access in <variant>. Include it first.
+#include <exception>
+
+#define C10_USING_CUSTOM_GENERATED_MACROS
+#define ET_ENABLE_DEPRECATED_CONSTANT_BUFFER 0
+#define FLATBUFFERS_MAX_ALIGNMENT 1024
+
+#include <executorch/runtime/platform/runtime.h>
+#include <executorch/runtime/executor/program.h>
+#include <executorch/runtime/executor/method.h>
+#include <executorch/runtime/core/memory_allocator.h>
+#include <executorch/runtime/executor/method_meta.h>
+#include <executorch/extension/data_loader/buffer_data_loader.h>
diff --git a/examples/arduino/README.md b/examples/arduino/README.md
new file mode 100644
index 00000000000..bd3a8d02f44
--- /dev/null
+++ b/examples/arduino/README.md
@@ -0,0 +1,211 @@
+<!---
+  Copyright (c) Meta Platforms, Inc. and affiliates.
+  All rights reserved.
+
+  This source code is licensed under the BSD-style license found in the
+  LICENSE file in the root directory of this source tree.
+--->
+
+# ExecuTorch Arduino Library
+
+Run PyTorch models on Arduino microcontrollers using ExecuTorch.
+
+This directory contains everything needed to package ExecuTorch as an
+Arduino library. A build script vendors the runtime sources from this
+repository into a self-contained library that Arduino users install
+through the Library Manager or by copying into their libraries folder.
+
+## How It Works
+
+```
+PyTorch Model ──► torch.export ──► .pte file ──► model.h (C array)
+                                                      │
+                                          Arduino Sketch (.ino)
+                                          #include <ExecuTorchArduino.h>
+                                          #include "model.h"
+                                                      │
+                                          arduino-cli compile ──► Upload ──► Runs on board
+```
+
+### The three pieces
+
+1. **The library** (`arduino_lib/ExecuTorchArduino/`) — the ExecuTorch
+   runtime, CMSIS-NN kernels, and portable ops packaged for the Arduino
+   build system.  Generated by `build_arduino_library.sh`; not checked in.
+
+2. **The model** (`model.h`) — a `.pte` file converted to a C byte array.
+   Each user brings their own model, exported from PyTorch with the
+   Cortex-M backend.
+
+3. **The sketch** (`.ino`) — a standard Arduino program that loads the
+   model, feeds it input, and reads the output.  Uses the native
+   ExecuTorch C++ API (`Program::load`, `Method::execute`, etc.).
+
+## Supported Boards
+
+| Board | MCU | Status |
+|-------|-----|--------|
+| Arduino Uno Q | STM32U585 (Cortex-M33) | Tested |
+| Arduino Nano 33 BLE | nRF52840 (Cortex-M4F) | Untested |
+| Arduino Giga R1 WiFi | STM32H747 (Cortex-M7) | Untested |
+| Arduino Portenta H7 | STM32H747 (Cortex-M7) | Untested |
+
+CMSIS-NN accelerated ops work on any board with an ARM Cortex-M processor
+with DSP extensions. Portable ops work on any architecture.
+
+## Quick Start
+
+### 1. Build the Arduino library
+
+```bash
+cd examples/arduino
+./build_arduino_library.sh
+```
+
+This copies the required ExecuTorch sources from the repository into
+`arduino_lib/ExecuTorchArduino/`, ready for Arduino.
+
+### 2. Install the library
+
+Copy the generated library into your Arduino libraries folder:
+
+```bash
+cp -r arduino_lib/ExecuTorchArduino ~/Arduino/libraries/
+```
+
+Or with `arduino-cli`:
+
+```bash
+arduino-cli lib install --zip-path arduino_lib/ExecuTorchArduino
+```
+
+### 3. Export a model
+
+Export a PyTorch model to `.pte` format with Cortex-M quantization, then
+convert to a C header:
+
+```bash
+python export_model.py --model my_model.pt --target cortex-m33 --output model.h
+```
+
+Or use one of the pre-exported models in the `examples/` directory.
+
+### 4. Write a sketch
+
+```cpp
+#include <ExecuTorchArduino.h>
+#include "model.h"
+
+using executorch::extension::BufferDataLoader;
+using executorch::runtime::Error;
+using executorch::runtime::HierarchicalAllocator;
+using executorch::runtime::MemoryAllocator;
+using executorch::runtime::MemoryManager;
+using executorch::runtime::Method;
+using executorch::runtime::MethodMeta;
+using executorch::runtime::Program;
+using executorch::runtime::Result;
+using executorch::runtime::Span;
+
+alignas(16) uint8_t method_pool[64 * 1024];
+alignas(16) uint8_t temp_pool[8 * 1024];
+
+void setup() {
+  Serial.begin(115200);
+  delay(2000);
+
+  executorch::runtime::runtime_init();
+
+  auto loader = BufferDataLoader(model_pte, model_pte_size);
+  Result<Program> program = Program::load(&loader);
+  if (!program.ok()) {
+    Serial.println("Failed to load program");
+    return;
+  }
+
+  // ... load method, set inputs, execute, read outputs
+  // See examples/ for complete working sketches.
+}
+
+void loop() {
+  // Run inference periodically
+  delay(2000);
+}
+```
+
+The sketch uses the **native ExecuTorch C++ API** — the same API used on
+Linux, Android, and bare-metal targets.  No wrapper layer, no
+Arduino-specific abstractions.
+
+### 5. Compile and upload
+
+```bash
+arduino-cli compile --fqbn arduino:zephyr:unoq MySketch
+arduino-cli upload  --fqbn arduino:zephyr:unoq -p /dev/ttyUSB0 MySketch
+```
+
+## What is inside the library
+
+The `build_arduino_library.sh` script assembles these components from
+the ExecuTorch repository:
+
+| Component | Source in repo | Purpose |
+|-----------|---------------|---------|
+| ET Runtime | `runtime/executor/`, `runtime/core/`, `runtime/kernel/`, `runtime/platform/` | Model loading, memory management, op dispatch |
+| Portable Ops | `kernels/portable/` | Software op implementations (any CPU) |
+| Cortex-M Ops | `backends/cortex_m/ops/` | CMSIS-NN accelerated int8 ops |
+| CMSIS-NN | fetched by cmake / Zephyr module | ARM's optimized DSP kernels |
+| flatcc | `third-party/flatcc/` | .pte file parsing |
+| flatbuffers | `third-party/flatbuffers/` | Schema headers |
+| c10 | `runtime/core/portable_type/c10/` | Core type definitions |
+
+The library uses no external dependencies beyond what the Arduino board
+core provides.
+
+## Arduino-specific patches
+
+The build script applies these patches to make ExecuTorch compile under
+Arduino's build system:
+
+1. **`#include <exception>` before `<variant>`** — Arduino's custom
+   `<new>` header omits `<exception>`, breaking `std::bad_variant_access`.
+
+2. **`cmake_macros.h` stub** — c10/torch headers expect a cmake-generated
+   file.  The stub defines `C10_USING_CUSTOM_GENERATED_MACROS`.
+
+3. **`platform_stubs.c`** — provides `_Exit()` and `fprintf()` for the
+   LLEXT environment on boards that lack them.
+
+## Development
+
+### Updating the library
+
+After modifying ExecuTorch sources, regenerate the library:
+
+```bash
+./build_arduino_library.sh        # rebuild
+./build_arduino_library.sh --clean  # remove generated output
+```
+
+### Testing
+
+```bash
+arduino-cli compile --fqbn arduino:zephyr:unoq examples/HelloExecuTorch
+arduino-cli upload  --fqbn arduino:zephyr:unoq -p /dev/ttyUSB0 examples/HelloExecuTorch
+arduino-cli monitor -p /dev/ttyUSB0 --config baudrate=115200
+```
+
+### Publishing to Arduino Library Manager
+
+The library is published by adding its repository URL to the
+[Arduino Library Registry](https://github.com/arduino/library-registry).
+After the initial registration, new git tags are picked up automatically.
+
+## Build Validation
+
+Tested on Arduino Uno Q (STM32U585, Cortex-M33 @ 160 MHz):
+
+- **390+ source files** compile with zero errors
+- **Flash**: 106 KB used (13% of 786 KB)
+- **RAM**: 91 KB used (69% of 131 KB)
+- **Model**: DS-CNN keyword spotting, int8 quantized via CMSIS-NN, 52.6 KB
diff --git a/examples/arduino/build_arduino_library.sh b/examples/arduino/build_arduino_library.sh
new file mode 100755
index 00000000000..505e121cf7b
--- /dev/null
+++ b/examples/arduino/build_arduino_library.sh
@@ -0,0 +1,236 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# build_arduino_library.sh — Assemble the ExecuTorch Arduino library
+# from sources in this repository.
+#
+# Usage:
+#   ./build_arduino_library.sh                # build the library
+#   ./build_arduino_library.sh --clean        # remove generated output
+#   ./build_arduino_library.sh --bump patch   # 0.1.0 → 0.1.1
+#   ./build_arduino_library.sh --bump minor   # 0.1.0 → 0.2.0
+#   ./build_arduino_library.sh --bump major   # 0.1.0 → 1.0.0
+#
+# Output: arduino_lib/ExecuTorchArduino/ (self-contained, installable)
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+ET_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+OUT_DIR="$SCRIPT_DIR/arduino_lib/ExecuTorchArduino"
+PROPS="$SCRIPT_DIR/library.properties"
+
+if [ "${1:-}" = "--clean" ]; then
+  echo "Cleaning generated library..."
+  rm -rf "$SCRIPT_DIR/arduino_lib"
+  echo "Done."
+  exit 0
+fi
+
+if [ "${1:-}" = "--bump" ]; then
+  PART="${2:-patch}"
+  CURRENT=$(grep "^version=" "$PROPS" | cut -d= -f2)
+  IFS='.' read -r MAJOR MINOR PATCH <<< "$CURRENT"
+  case "$PART" in
+    major) MAJOR=$((MAJOR + 1)); MINOR=0; PATCH=0 ;;
+    minor) MINOR=$((MINOR + 1)); PATCH=0 ;;
+    patch) PATCH=$((PATCH + 1)) ;;
+    *) echo "Usage: $0 --bump [major|minor|patch]"; exit 1 ;;
+  esac
+  NEW="$MAJOR.$MINOR.$PATCH"
+  sed -i '' "s/^version=.*/version=$NEW/" "$PROPS" 2>/dev/null || \
+    sed -i "s/^version=.*/version=$NEW/" "$PROPS"
+  echo "Version: $CURRENT → $NEW"
+  exit 0
+fi
+
+echo "=== Building ExecuTorch Arduino Library ==="
+echo "  ET repo:  $ET_ROOT"
+echo "  Output:   $OUT_DIR"
+
+rm -rf "$OUT_DIR"
+mkdir -p "$OUT_DIR/src" "$OUT_DIR/examples"
+
+# ─────────────────────────────────────────────────────────
+# 1. Copy library metadata, wrapper header, and stubs
+# ─────────────────────────────────────────────────────────
+cp "$SCRIPT_DIR/library.properties" "$OUT_DIR/"
+cp "$SCRIPT_DIR/ExecuTorchArduino.h" "$OUT_DIR/src/"
+cp "$SCRIPT_DIR/platform_stubs.c" "$OUT_DIR/src/"
+cp -r "$SCRIPT_DIR/examples/"* "$OUT_DIR/examples/"
+
+echo "[1/7] Metadata and examples copied"
+
+# ─────────────────────────────────────────────────────────
+# 2. Vendor ET runtime sources
+# ─────────────────────────────────────────────────────────
+ET_SRC="$OUT_DIR/src/executorch"
+mkdir -p "$ET_SRC"
+
+for dir in runtime/core runtime/executor runtime/kernel \
+           runtime/platform runtime/backend; do
+  mkdir -p "$ET_SRC/$dir"
+  find "$ET_ROOT/$dir" -maxdepth 1 -name "*.h" -exec cp {} "$ET_SRC/$dir/" \;
+  find "$ET_ROOT/$dir" -maxdepth 1 -name "*.cpp" -exec cp {} "$ET_SRC/$dir/" \;
+done
+
+# Subdirectories with headers
+for dir in runtime/core/exec_aten runtime/core/exec_aten/util \
+           runtime/core/portable_type runtime/platform/default; do
+  mkdir -p "$ET_SRC/$dir"
+  find "$ET_ROOT/$dir" -maxdepth 1 -name "*.h" -exec cp {} "$ET_SRC/$dir/" \;
+  find "$ET_ROOT/$dir" -maxdepth 1 -name "*.cpp" -exec cp {} "$ET_SRC/$dir/" \;
+done
+
+# Extension data loader
+mkdir -p "$ET_SRC/extension/data_loader"
+cp "$ET_ROOT/extension/data_loader/"*.h "$ET_SRC/extension/data_loader/"
+cp "$ET_ROOT/extension/data_loader/buffer_data_loader.cpp" \
+   "$ET_SRC/extension/data_loader/" 2>/dev/null || true
+
+# Runner util headers
+mkdir -p "$ET_SRC/extension/runner_util"
+cp "$ET_ROOT/extension/runner_util/"*.h "$ET_SRC/extension/runner_util/" 2>/dev/null || true
+
+# Schema headers (generated — need a prior cmake build)
+mkdir -p "$ET_SRC/schema"
+cp "$ET_ROOT/schema/"*.h "$ET_SRC/schema/" 2>/dev/null || true
+# Look for generated headers in common build dirs
+for build_dir in "$ET_ROOT/cmake-out" "$ET_ROOT/cmake-out-mac" \
+                 "$ET_ROOT/outputs/build_uno_q"; do
+  if [ -d "$build_dir/schema/include/executorch/schema" ]; then
+    cp "$build_dir/schema/include/executorch/schema/"*.h "$ET_SRC/schema/"
+    break
+  fi
+done
+
+echo "[2/7] ET runtime sources copied"
+
+# ─────────────────────────────────────────────────────────
+# 3. Vendor portable kernels
+# ─────────────────────────────────────────────────────────
+mkdir -p "$ET_SRC/kernels/portable/cpu/util" \
+         "$ET_SRC/kernels/portable/cpu/pattern"
+
+# Copy all portable op sources and headers
+find "$ET_ROOT/kernels/portable/cpu" -maxdepth 1 \( -name "*.cpp" -o -name "*.h" \) \
+  -exec cp {} "$ET_SRC/kernels/portable/cpu/" \;
+cp "$ET_ROOT/kernels/portable/cpu/util/"*.h "$ET_SRC/kernels/portable/cpu/util/"
+cp "$ET_ROOT/kernels/portable/cpu/util/"*.cpp "$ET_SRC/kernels/portable/cpu/util/" 2>/dev/null || true
+cp "$ET_ROOT/kernels/portable/cpu/pattern/"*.h "$ET_SRC/kernels/portable/cpu/pattern/"
+cp "$ET_ROOT/kernels/portable/cpu/pattern/"*.cpp "$ET_SRC/kernels/portable/cpu/pattern/" 2>/dev/null || true
+
+echo "[3/7] Portable kernels copied"
+
+# ─────────────────────────────────────────────────────────
+# 4. Vendor Cortex-M backend ops
+# ─────────────────────────────────────────────────────────
+mkdir -p "$ET_SRC/backends/cortex_m/ops"
+cp "$ET_ROOT/backends/cortex_m/ops/"*.cpp "$ET_SRC/backends/cortex_m/ops/"
+cp "$ET_ROOT/backends/cortex_m/ops/"*.h "$ET_SRC/backends/cortex_m/ops/"
+
+echo "[4/7] Cortex-M ops copied"
+
+# ─────────────────────────────────────────────────────────
+# 5. Vendor third-party dependencies
+# ─────────────────────────────────────────────────────────
+
+# c10 / torch headers
+cp -r "$ET_ROOT/runtime/core/portable_type/c10/c10" "$OUT_DIR/src/c10"
+cp -r "$ET_ROOT/runtime/core/portable_type/c10/torch" "$OUT_DIR/src/torch"
+
+# cmake_macros.h stub
+mkdir -p "$OUT_DIR/src/torch/headeronly/macros"
+cat > "$OUT_DIR/src/torch/headeronly/macros/cmake_macros.h" << 'STUB'
+#pragma once
+#define C10_BUILD_SHARED_LIBS
+#define C10_USE_GLOG 0
+#define C10_USE_MINIMAL_GLOG 0
+#define C10_USE_GFLAGS 0
+STUB
+
+# flatcc runtime
+mkdir -p "$OUT_DIR/src/flatcc"
+cp -r "$ET_ROOT/third-party/flatcc/include/flatcc" "$OUT_DIR/src/flatcc/"
+mkdir -p "$OUT_DIR/src/flatcc/runtime"
+cp "$ET_ROOT/third-party/flatcc/src/runtime/"*.c "$OUT_DIR/src/flatcc/runtime/"
+
+# flatbuffers headers
+cp -r "$ET_ROOT/third-party/flatbuffers/include/flatbuffers" "$OUT_DIR/src/flatbuffers"
+
+# CMSIS-NN (from Zephyr workspace or cmake fetchcontent)
+CMSIS_NN=""
+for candidate in \
+  "$ET_ROOT/outputs/zephyrproject/modules/lib/cmsis-nn" \
+  "$ET_ROOT/third-party/cmsis-nn" \
+  "$ET_ROOT/backends/arm/third-party/cmsis-nn/CMSIS-NN"; do
+  if [ -d "$candidate/Source" ]; then
+    CMSIS_NN="$candidate"
+    break
+  fi
+done
+
+if [ -n "$CMSIS_NN" ]; then
+  mkdir -p "$OUT_DIR/src/cmsis-nn"
+  cp -r "$CMSIS_NN/Source" "$OUT_DIR/src/cmsis-nn/"
+  cp "$CMSIS_NN/Include/"*.h "$OUT_DIR/src/" 2>/dev/null || true
+  if [ -d "$CMSIS_NN/Include/Internal" ]; then
+    mkdir -p "$OUT_DIR/src/Internal"
+    cp "$CMSIS_NN/Include/Internal/"*.h "$OUT_DIR/src/Internal/"
+  fi
+  echo "[5/7] CMSIS-NN copied from $CMSIS_NN"
+else
+  echo "[5/7] WARNING: CMSIS-NN not found. Cortex-M ops will not link."
+fi
+
+# CMSIS Core headers (for arm_math_types.h)
+for candidate in \
+  "$ET_ROOT/outputs/zephyrproject/modules/hal/cmsis_6/CMSIS/Core/Include" \
+  "$ET_ROOT/third-party/cmsis/CMSIS/Core/Include"; do
+  if [ -d "$candidate" ]; then
+    cp "$candidate/"*.h "$OUT_DIR/src/" 2>/dev/null || true
+    break
+  fi
+done
+
+echo "[6/7] Third-party dependencies copied"
+
+# ─────────────────────────────────────────────────────────
+# 6. Apply Arduino-specific patches
+# ─────────────────────────────────────────────────────────
+
+# Fix: #include <exception> before <variant> in all ET headers
+find "$OUT_DIR/src/executorch" -name "*.h" -exec \
+  grep -l '#include <variant>' {} \; | while read f; do
+  sed -i '' 's|#include <variant>|#include <exception>\n#include <variant>|' "$f" 2>/dev/null || \
+  sed -i 's|#include <variant>|#include <exception>\n#include <variant>|' "$f"
+done
+
+# Remove test files, ATen-specific files
+find "$OUT_DIR" -path "*testing*" -delete 2>/dev/null || true
+find "$OUT_DIR" -name "*_aten.cpp" -delete 2>/dev/null || true
+find "$OUT_DIR" -path "*test*" -name "*.cpp" -delete 2>/dev/null || true
+
+echo "[7/7] Arduino patches applied"
+
+# ─────────────────────────────────────────────────────────
+# Summary
+# ─────────────────────────────────────────────────────────
+NSRC=$(find "$OUT_DIR/src" -name "*.cpp" -o -name "*.c" | wc -l | tr -d ' ')
+NHDR=$(find "$OUT_DIR/src" -name "*.h" | wc -l | tr -d ' ')
+
+echo ""
+echo "=== Library built ==="
+echo "  Location:  $OUT_DIR"
+echo "  Sources:   $NSRC"
+echo "  Headers:   $NHDR"
+echo ""
+echo "Install:"
+echo "  cp -r $OUT_DIR ~/Arduino/libraries/"
+echo ""
+echo "Or clean up:"
+echo "  $0 --clean"
diff --git a/examples/arduino/examples/HelloExecuTorch/HelloExecuTorch.ino b/examples/arduino/examples/HelloExecuTorch/HelloExecuTorch.ino
new file mode 100644
index 00000000000..b2502152505
--- /dev/null
+++ b/examples/arduino/examples/HelloExecuTorch/HelloExecuTorch.ino
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// HelloExecuTorch — Minimal ExecuTorch sketch
+//
+// Initializes the ExecuTorch runtime and loads a model using the core
+// ET library (portable ops only, no hardware-specific backends).
+// Use this to verify the library works on your board.
+
+#include <ExecuTorchArduino.h>
+#include "model.h"
+
+using executorch::extension::BufferDataLoader;
+using executorch::runtime::MemoryAllocator;
+using executorch::runtime::Program;
+using executorch::runtime::Result;
+
+alignas(16) static uint8_t arena[32 * 1024];
+
+void setup() {
+  Serial.begin(115200);
+  delay(2000);
+
+  Serial.println("=== HelloExecuTorch ===");
+
+  executorch::runtime::runtime_init();
+  Serial.println("Runtime initialized.");
+
+  auto loader = BufferDataLoader(model_pte, model_pte_size);
+  Result<Program> program = Program::load(&loader);
+  if (program.ok()) {
+    Serial.println("Model loaded OK!");
+    Serial.print("  Size: ");
+    Serial.print(model_pte_size);
+    Serial.println(" bytes");
+    Serial.print("  Methods: ");
+    Serial.println(program->num_methods());
+  } else {
+    Serial.println("ERROR: Model load failed");
+  }
+}
+
+void loop() {
+  Serial.println("ExecuTorch ready");
+  delay(5000);
+}
diff --git a/examples/arduino/examples/KeywordSpotting/KeywordSpotting.ino b/examples/arduino/examples/KeywordSpotting/KeywordSpotting.ino
new file mode 100644
index 00000000000..3e8f48c9c32
--- /dev/null
+++ b/examples/arduino/examples/KeywordSpotting/KeywordSpotting.ino
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// KeywordSpotting — DS-CNN inference with ExecuTorch + CMSIS-NN
+//
+// Runs a quantized DS-CNN model (MLPerf Tiny KWS benchmark) using the
+// Cortex-M backend with CMSIS-NN accelerated int8 ops.  The model
+// classifies 12 keywords: silence, unknown, yes, no, up, down, left,
+// right, on, off, stop, go.
+//
+// This example demonstrates:
+//   - Loading a .pte model with cortex_m::* ops (CMSIS-NN backend)
+//   - Setting up memory allocators for constrained devices
+//   - Running inference and reading output tensors
+//
+// Required files (not included, generate with export_model.py):
+//   - model.h       : DS-CNN .pte exported with CortexMQuantizer
+//   - test_inputs.h : MFCC features from real audio samples
+
+#include <ExecuTorchArduino.h>
+#include "model.h"
+
+using executorch::aten::ScalarType;
+using executorch::extension::BufferDataLoader;
+using executorch::runtime::Error;
+using executorch::runtime::EValue;
+using executorch::runtime::HierarchicalAllocator;
+using executorch::runtime::MemoryAllocator;
+using executorch::runtime::MemoryManager;
+using executorch::runtime::Method;
+using executorch::runtime::MethodMeta;
+using executorch::runtime::Program;
+using executorch::runtime::Result;
+using executorch::runtime::Span;
+
+static const char* kLabels[] = {
+    "silence", "unknown", "yes", "no", "up", "down",
+    "left", "right", "on", "off", "stop", "go"};
+static constexpr int kNumClasses = 12;
+
+alignas(16) static uint8_t method_pool[64 * 1024];
+alignas(16) static uint8_t temp_pool[8 * 1024];
+
+void setup() {
+  Serial.begin(115200);
+  delay(2000);
+
+  Serial.println("=== ExecuTorch Keyword Spotting ===");
+  Serial.print("Model: ");
+  Serial.print(model_pte_size);
+  Serial.println(" bytes (DS-CNN int8, CMSIS-NN)");
+
+  executorch::runtime::runtime_init();
+  Serial.println("Runtime initialized.");
+}
+
+void loop() {
+  auto loader = BufferDataLoader(model_pte, model_pte_size);
+  Result<Program> program = Program::load(&loader);
+  if (!program.ok()) {
+    Serial.println("ERROR: Program::load failed");
+    delay(5000);
+    return;
+  }
+
+  const char* method_name = nullptr;
+  {
+    auto r = program->get_method_name(0);
+    if (!r.ok()) {
+      Serial.println("ERROR: no methods");
+      delay(5000);
+      return;
+    }
+    method_name = *r;
+  }
+
+  auto meta = program->method_meta(method_name);
+  if (!meta.ok()) {
+    Serial.println("ERROR: method_meta failed");
+    delay(5000);
+    return;
+  }
+
+  MemoryAllocator method_allocator(sizeof(method_pool), method_pool);
+
+  size_t n_planned = meta->num_memory_planned_buffers();
+  Span<uint8_t>* spans = static_cast<Span<uint8_t>*>(
+      method_allocator.allocate(n_planned * sizeof(Span<uint8_t>)));
+  for (size_t i = 0; i < n_planned; i++) {
+    size_t sz = static_cast<size_t>(meta->memory_planned_buffer_size(i).get());
+    uint8_t* buf = static_cast<uint8_t*>(method_allocator.allocate(sz));
+    spans[i] = {buf, sz};
+  }
+
+  HierarchicalAllocator planned({spans, n_planned});
+  MemoryAllocator temp_allocator(sizeof(temp_pool), temp_pool);
+  MemoryManager mm(&method_allocator, &planned, &temp_allocator);
+
+  auto method = program->load_method(method_name, &mm);
+  if (!method.ok()) {
+    Serial.print("ERROR: load_method: 0x");
+    Serial.println((int)method.error(), HEX);
+    delay(5000);
+    return;
+  }
+
+  Error status = method->execute();
+  if (status != Error::Ok) {
+    Serial.print("ERROR: execute: 0x");
+    Serial.println((int)status, HEX);
+    delay(5000);
+    return;
+  }
+
+  EValue output;
+  method->get_outputs(&output, 1);
+  if (output.isTensor()) {
+    auto tensor = output.toTensor();
+    int best = 0;
+    float best_val = -1e9f;
+    for (int i = 0; i < kNumClasses && i < tensor.numel(); i++) {
+      float val = (tensor.scalar_type() == ScalarType::Float)
+          ? tensor.const_data_ptr<float>()[i]
+          : static_cast<float>(tensor.const_data_ptr<int8_t>()[i]);
+      if (val > best_val) {
+        best_val = val;
+        best = i;
+      }
+    }
+    Serial.print("Detected: ");
+    Serial.println(kLabels[best]);
+  }
+
+  delay(5000);
+}
diff --git a/examples/arduino/library.properties b/examples/arduino/library.properties
new file mode 100644
index 00000000000..5288800f27a
--- /dev/null
+++ b/examples/arduino/library.properties
@@ -0,0 +1,16 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+name=ExecuTorchArduino
+version=0.1.0
+author=Meta Platforms
+maintainer=ExecuTorch Team <executorch@meta.com>
+sentence=Run PyTorch models on Arduino microcontrollers with ExecuTorch.
+paragraph=ExecuTorch is a PyTorch runtime optimized for on-device inference. This library packages the ExecuTorch runtime with CMSIS-NN acceleration for ARM Cortex-M boards, enabling quantized model inference directly from Arduino sketches.
+category=Data Processing
+url=https://github.com/pytorch/executorch
+architectures=zephyr,mbed_nano,mbed_giga,mbed_portenta
+includes=ExecuTorchArduino.h
diff --git a/examples/arduino/platform_stubs.c b/examples/arduino/platform_stubs.c
new file mode 100644
index 00000000000..6d50ed3290c
--- /dev/null
+++ b/examples/arduino/platform_stubs.c
@@ -0,0 +1,25 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// Platform stubs for Arduino LLEXT environment.
+// These C library functions are not exported by the Zephyr kernel's
+// LLEXT symbol table but are referenced by the ExecuTorch runtime.
+
+#include <stdarg.h>
+#include <stdio.h>
+
+void _Exit(int status) {
+    (void)status;
+    while (1) {}
+}
+
+int fprintf(FILE* stream, const char* fmt, ...) {
+    (void)stream;
+    (void)fmt;
+    return 0;
+}
diff --git a/zephyr/samples/hello-executorch/boards/arduino_uno_q.conf b/zephyr/samples/hello-executorch/boards/arduino_uno_q.conf
new file mode 100644
index 00000000000..d3bb1935979
--- /dev/null
+++ b/zephyr/samples/hello-executorch/boards/arduino_uno_q.conf
@@ -0,0 +1,20 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# Arduino Uno Q — STM32U585 (Cortex-M33, DSP, no MVE, no Ethos-U)
+# 2 MB flash, 786 KB SRAM — plenty of room for DS-CNN int8 (50 KB .pte).
+#
+# CMSIS-NN uses the DSP path for M33 (no MVE on U5 series).
+
+# Memory tuning for DS-CNN on STM32U585:
+#   - method allocator: 64 KB (DS-CNN metadata + planned buffers)
+#   - temp allocator:   8 KB  (scratch for CMSIS-NN kernels)
+#   - main stack:       8 KB
+#   - heap:            16 KB
+CONFIG_EXECUTORCH_METHOD_ALLOCATOR_POOL_SIZE=65536
+CONFIG_EXECUTORCH_TEMP_ALLOCATOR_POOL_SIZE=8192
+CONFIG_MAIN_STACK_SIZE=8192
+CONFIG_HEAP_MEM_POOL_SIZE=16384