// the find
sipeed/MaixPy-v1
MicroPython for K210 RISC-V, let's play with edge AI easier
MicroPython firmware for the Kendryte K210, a RISC-V chip with a built-in neural network accelerator (KPU). It targets Sipeed's Maix boards and lets you do camera capture, LCD display, and on-device CNN inference (face detection, object recognition) from Python instead of C. Good fit for hobbyists and makers doing embedded vision who don't want to write bare-metal C against the Kendryte SDK.
The API is genuinely simple — `sensor.snapshot()` + `kpu.forward()` gets you a running classifier in under 10 lines, which is rare for anything touching a hardware NPU. It doubles as a C SDK (MicroPython can be excluded from the build) if you outgrow the Python layer but want to keep the board support and drivers. Driver coverage is broad for a chip this niche: SD card, ws2812, multiple LCD controllers, I2S mic array, even an NES emulator port, so peripheral support isn't the bottleneck it usually is on obscure silicon.
This is explicitly the 'v1' / legacy repo — Sipeed's `MaixPy` is the maintained successor, and the historic branch here is dead weight kept only for commit history. The KPU/nncase inference libraries ship as precompiled `.a` blobs in `components/kendryte_sdk/libs`, not source, so you're trusting vendor binaries for the one part of the stack most people are here for. Building from source requires a specific Kendryte toolchain tarball from Sipeed's own CDN plus several git submodules (micropython core, lv_bindings, kendryte-standalone-sdk) — not a `pip install` experience, and docs live on a separate site rather than in-repo. The K210 itself is an aging chip with no official long-term roadmap from Kendryte, so this is a good weekend-project platform but a risky choice for anything you plan to ship for years.