Kimi CLI on Smart Pi One¶

Kimi CLI is Moonshot AI's official Kimi command-line assistant, packaged by YUMI-LAB to run on the 32-bit ARM (Allwinner H3 / armv7l) Smart Pi One and Smart Pad — hardware Moonshot ships no 32-bit build for.

Repository: github.com/Yumi-Lab/kimi-cli-smartpi

1. How it works¶

Moonshot's official installer has no 32-bit ARM build, but the kimi-cli PyPI package ships as a pure-Python wheel (py3-none-any). YUMI-LAB installs it with uv (uv tool install kimi-cli), so Kimi CLI runs natively — no emulation and no version pinning. Only a couple of C dependencies (Pillow) compile from source with the armhf toolchain during install; earlyoom is added for memory safety on the 1 GB board.

2. Requirements¶

armv7l / 32-bit ARM SBC with at least 1 GB RAM
A Debian-based Linux distribution (tested on the Smart Pad — Debian 13 trixie armhf, Python 3.13)
~/.local/bin on your $PATH
A Kimi account (Moonshot AI) — sign in with OAuth or an API key

3. Installation¶

Run the one-line installer on your Smart Pi One as a normal user (not root):

curl -fsSL https://raw.githubusercontent.com/Yumi-Lab/kimi-cli-smartpi/main/install.sh | bash

The first install compiles Pillow from source (a few minutes; 4 cores by default). The installer is idempotent — re-running it skips work that's already done.

Fanless boards¶

On a board with no active cooling, cap the compile to 2 cores to avoid thermal throttling (peaks ~88 °C instead of ~102 °C at 4 cores):

KIMI_BUILD_CPUS=0,1 curl -fsSL https://raw.githubusercontent.com/Yumi-Lab/kimi-cli-smartpi/main/install.sh | bash

4. Authentication¶

kimi login

Follow the OAuth prompt, or provide an API key. Credentials are stored in ~/.kimi/.

5. Usage¶

Command	Purpose
`kimi`	Full interactive agent TUI (native execution)
`kimi -p "your question"`	Start from a prompt, then continue interactively
`kimi --quiet -p "your task"`	One-shot mode — final answer only
`kimi login`	Sign in to your Kimi account
`kimi --version`	Version check (~0.9 s)

In agent mode, Kimi CLI can read and edit files and run commands on the board — useful for scripting the GPIO sensors, configuring services, or working through Smart Pi projects.

Limiting runtime CPU cores¶

Use KIMI_CPUS to pin the agent to specific cores at runtime (the counterpart of GROK_CPUS), keeping the board cool and responsive:

KIMI_CPUS=0,1 kimi -p "explain this error"

Example Kimi CLI session on a Smart Pi One Example session (illustration).

6. Notes¶

Native, no emulation: unlike Grok CLI (which runs under QEMU), Kimi CLI runs directly as a Python tool. Startup is fast (~0.9 s); one-shot latency is bounded by Moonshot's remote inference, not the board.
Updating Kimi: do not re-run uv tool install kimi-cli over an existing install. Update with uv tool upgrade kimi-cli, then re-run the installer once to restore the YUMI-LAB wrappers.
CPU control: KIMI_BUILD_CPUS bounds the one-time compile; KIMI_CPUS bounds the running agent. Both default to 4 cores.
Licensing: the installer scripts are MIT (YUMI-LAB); Kimi CLI itself remains subject to Moonshot AI's terms and is installed from the official PyPI package at runtime.