maixpy-dev - SKILL.md Agent Skill

name: maixpy-dev description: | MaixPy v4 development assistant for Sipeed MaixCAM, MaixCAM-Pro, and MaixCAM2 edge AI devices. Use this skill when developing Python applications for MaixPy platform including: (1) AI vision applications (YOLO detection, classification, face recognition, pose estimation) (2) Image processing (find blobs, edges, QR codes, barcodes, lines) (3) Hardware peripherals (camera, display, UART, I2C, SPI, GPIO, PWM, ADC, USB HID) (4) Network applications (WiFi, HTTP streaming, MQTT, WebSocket, RTSP/RTMP) (5) Audio applications (playback, recording, TTS, ASR) (6) LLM/VLM integration (Qwen, DeepSeek, InternVL on MaixCAM2) (7) Object tracking and counting (ByteTracker) (8) Video encoding/decoding, OpenCV integration Trigger when user mentions MaixPy, MaixCAM, MaixCAM2, MaixCAM-Pro, Sipeed AI development, or embedded AI vision projects.

MaixPy Development Skill

MaixPy v4 is a Python SDK for edge AI development on Sipeed hardware. This skill provides patterns, examples, and best practices.

Hardware Comparison

Feature	MaixCAM/MaixCAM-Pro	MaixCAM2
CPU	1GHz RISC-V (Linux)	1.2GHz A53 x2 (Ubuntu)
Memory	256MB DDR3	1GB/4GB LPDDR4
NPU	1Tops@INT8	3.2Tops@INT8
LLM Support	No	Yes (Qwen/DeepSeek)

Quick Start

from maix import camera, display, image, nn, app

detector = nn.YOLOv8(model="/root/models/yolov8n.mud", dual_buff=True)
cam = camera.Camera(detector.input_width(), detector.input_height(), detector.input_format())
disp = display.Display()

while not app.need_exit():
    img = cam.read()
    objs = detector.detect(img, conf_th=0.5, iou_th=0.45)
    for obj in objs:
        img.draw_rect(obj.x, obj.y, obj.w, obj.h, color=image.COLOR_RED)
        img.draw_string(obj.x, obj.y, f'{detector.labels[obj.class_id]}: {obj.score:.2f}')
    disp.show(img)

Reference Documentation

Topic	File	Content
AI Models	ai_models.md	YOLO, classifier, face, OCR, pose, segmentation
Image Processing	image_processing.md	Draw, blobs, edges, QR/barcodes, transforms
Peripherals	peripherals.md	UART, I2C, SPI, GPIO, PWM, ADC
Network	network.md	WiFi, HTTP, MQTT, WebSocket
Audio	audio.md	Playback, recording, TTS, ASR
LLM/VLM	llm_vlm.md	Qwen, DeepSeek, InternVL (MaixCAM2)
Tracking	tracking.md	ByteTracker, counting, trajectories
Patterns	patterns.md	Touch UI, threading, state machine, i18n
Advanced	advanced.md	OpenCV, video, USB HID, RTSP, protocols

Device Detection

from maix import sys

device_id = sys.device_id()  # "maixcam", "maixcam2"

if device_id == "maixcam2":
    model = "/root/models/yolo11s.mud"  # Larger model
else:
    model = "/root/models/yolov8n.mud"  # Nano model

Model Paths

Pre-installed models in /root/models/:

Detection: yolov8n.mud, yolo11n.mud, yolo11s.mud
Segmentation: yolo11n_seg.mud, yolov8n_seg.mud
Pose: yolo11n_pose.mud, yolov8n_pose.mud
Face: yolov8n_face.mud, retinaface.mud
Classifier: mobilenetv2.mud
OCR: pp_ocr.mud
Hand: hand_landmarks.mud

App Development

my_app/
├── app.yaml      # Config (see assets/app.yaml.template)
├── main.py       # Entry point
└── icon.png      # App icon (128x128)

Resources

Docs: https://wiki.sipeed.com/maixpy/
API: https://wiki.sipeed.com/maixpy/api/index.html
GitHub: https://github.com/sipeed/MaixPy
Examples: https://github.com/sipeed/MaixPy/tree/main/examples
Projects: https://github.com/sipeed/MaixPy/tree/main/projects
MaixHub: https://maixhub.com
Community: QQ群 862340358, t.me/maixpy