ShiMetaPi Workbench is a desktop IDE for users of the ShiMetaPi Pico-G1 (GK7206) development board, powered by Claude Code. It brings the whole workflow — "account login → browse models → one-click deploy to the board → write Python in the IDE / view the camera feed / let AI debug the board" — into a single client.
After training a model on the AIHub training platform, Pico-G1 users traditionally had to jump across multiple tools:
- Log in to AIHub in a browser and download
.xmm.zip - Push it to the board's SD card with
scp - SSH into the board and manually run
shimeta-svp / shimeta-xmm / shimeta-py - Open a WebUI in a browser or pull RTSP in VLC to see the picture
The whole process spans 4–5 tools and 10+ manual operations. Workbench integrates 6 top Tabs + three on-board tools, so users can do everything in one window.
| User type | Typical scenario |
|---|
| AI app developers | Trained a model on AIHub, want to quickly validate it on Pico-G1 |
| Embedded engineers | Run custom Python scripts on the board, debug hardware peripherals |
| AI Agent explorers | Operate the board with Claude Code in natural language |
| Solution integrators | Demonstrate 9 scenario-based detections (person/vehicle/fire, etc.) |
| Module | Form | Role |
|---|
| Workbench IDE | Windows desktop app | One-stop entry via 6 top Tabs |
| AIHub platform | Cloud | Model training and model repository |
| Board-side tools | Pushed over SSH | shimeta-py (Python runtime) / shimeta-svp (9 scenarios) / shimeta-xmm (custom models) |
| Tab | Main capability |
|---|
| Agent | Claude Code natural-language debugging of the board |
| shimeta-py IDE | Python code editor, runs directly on the board |
| Terminal | SSH terminal connecting to the board |
| Skills | Skill package management |
| Models | Model Market + user models, one-click deploy to the board |
| Vision | Launch shimeta-svp / shimeta-xmm, view the camera feed on the right |
┌─────────────────────────────────────────────────────────────┐
│ AIHub Model Training Platform (cloud) aihub.shimetapi.cn │
│ ─ Upload data → annotate → train → export model │
│ ─ Docs: software-platform/model-finetuning/ │
└────────────────────────┬────────────────────────────────────┘
│ HTTPS / model training link
▼
┌─────────────────────────────────────────────────────────────┐
│ Workbench (Windows desktop IDE) │
│ ─ 6 Tabs: Agent / IDE / Terminal / Skills / Models / Vision│
│ ─ Claude Code Agent drives board-side debugging │
│ ─ Models Tab deploys to the board in one click │
└────────────────────────┬────────────────────────────────────┘
│ SSH + scp
▼
┌─────────────────────────────────────────────────────────────┐
│ Pico-G1 board (GK7206) 192.168.49.10 │
│ ─ shimeta-py v1.0.4 Python runtime │
│ ─ shimeta-svp 9 scenario-based detections │
│ ─ shimeta-xmm custom model inference │
│ ─ user shimetad (passwordless SSH) / root/123456 │
└─────────────────────────────────────────────────────────────┘
| Product | Documentation | Role |
|---|
| ShiMetaPi AIHub | software-platform/model-finetuning/ | Cloud training and model repository |
| ShiMetaPi Workbench | software-platform/workbench/ (this doc) | Desktop IDE |
| ShimetaPi Repository | software-platform/shimetapi-repo/ | Board-side CLI tools and distribution |
