AI on Smart Pi One¶
The Smart Pi One and Smart Pad are 32-bit ARM (Allwinner H3 / armv7l) single-board computers. Most modern AI command-line tools ship only for 64-bit machines and refuse to install on this architecture. YUMI-LAB maintains dedicated installers that bring today's leading AI coding assistants to the board — running an interactive AI agent directly in the terminal, on hardware with as little as 1 GB of RAM.
1. Available AI Assistants¶
| Tool | Provider | Sign-in | Guide |
|---|---|---|---|
| Anthropic | Claude Pro or Max account (no API key) | Claude Code | |
| xAI | grok.com / SuperGrok account (no API key) | Grok CLI | |
| Mistral AI | Mistral API key | Vibe CLI | |
| Moonshot AI | Kimi account (OAuth or API key) | Kimi CLI |
Claude Code and Grok CLI sign in with your existing subscription account through a browser-based flow — no API key, no separate developer billing. Kimi CLI signs in with a Kimi account (OAuth or an API key). Vibe CLI uses a Mistral API key (created at console.mistral.ai).
2. Requirements¶
- A Smart Pi One or Smart Pad running a Debian-based Linux (see Install Linux)
armv7l/ 32-bit ARM CPU (Allwinner H3)- At least 1 GB RAM
- An active internet connection
- Access to the matching provider: a Claude Pro/Max account (Claude Code), a grok.com / SuperGrok account (Grok CLI), a Mistral API key (Vibe CLI), or a Kimi account (Kimi CLI)
3. What can you do with it?¶
These assistants run in agent mode: they can read and write files, run shell commands, and help you build and debug projects directly on the board. Typical uses on a Smart Pi One:
- Writing and fixing Python scripts for the GPIO sensors and modules
- Setting up services (Klipper, Home Assistant, media servers…)
- Explaining Linux commands and troubleshooting the system
- Prototyping small projects with a full conversational coding assistant
4. Which one should I choose?¶
All three give you an interactive terminal UI and one-shot (-p "question") query mode. Pick the one that matches the account or key you already have:
- Choose Claude Code if you have a Claude Pro or Max account. It runs natively (no emulation) and is the fastest.
- Choose Grok CLI if you have a grok.com / SuperGrok account. It runs the official binary under lightweight QEMU emulation.
- Choose Vibe CLI if you have a Mistral API key. It's a thin client — inference runs on Mistral's servers, so it's light on the board (but the first install compiles for ~15 min).
- Choose Kimi CLI if you have a Kimi account. It installs natively as a pure-Python tool and starts fast (~0.9 s).
Nothing stops you from installing all of them.