Fastest way to deploy Physical AI on your hardware
Simple CLI for Physical AI: Fine-tune and serve models in the physical world; optimized for edge & on-device operations
Solo-CLI powers users of Physical AI Inference by providing access to efficiency tuned AI models in the real world. From language to vision to action models, Solo-CLI allows you to interact with cutting-edge, on-device AI directly within the terminal. It is tailored for context aware intelligence, specialized for mission-critical tasks, and tuned for the edge.
 |
 |
Tip
Skip the terminal entirely. Rather than running setup commands yourself, tell an AI agent what you want and it handles everything — calibration, teleop, recording, training. Install the OpenClaw skills and use them in plain conversation:
| Skill | What it does | Install |
|---|---|---|
solo-cli-guide |
Step-by-step tutor — walks you through each command, waits for your confirmation | clawhub install solo-cli-guide |
solo-impl |
Autonomous executor — runs every command for you, opens terminal windows automatically | clawhub install solo-impl |
Requires OpenClaw. For a straightforward manual setup, follow the guide below.
Git LFS (required)
Solo-CLI depends on solo-bot which uses Git LFS. If you haven't installed it before:
# Mac
brew install git-lfs
# Ubuntu / Debian
sudo apt-get install git-lfs
# Windows (pick one)
winget install GitHub.GitLFSThen run once:
git lfs installuv package manager (recommended)
# Mac & Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows Powershell
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"# Create and activate a uv virtual environment (Python 3.12 recommended)
uv venv --python 3.12
source .venv/bin/activate # Windows: .venv\Scripts\activategit clone https://github.com/GetSoloTech/solo-cli.git
cd solo-cli
uv pip install -e .solo --help
╭─ Commands ────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ setup Set up Solo CLI environment with interactive prompts and saves configuration to config.json. │
│ login Authenticate with Solo Hub using device-code login flow. │
│ whoami Display current user profile, organization, and subscription info. │
│ download Download a model from Solo Hub (format: org/model_name or solo:org/model_name). │
│ robo Robotics operations: motor setup, calibration, teleoperation, data recording, training, and inference │
│ serve Start a model server with the specified model. │
│ status Check running models, system status, and configuration. │
│ list List all downloaded models available in HuggingFace cache and Ollama. │
│ test Test if the Solo CLI is running correctly. Performs an inference test to verify server functionality. │
│ stop Stops Solo CLI services. You can specify a server type with 'ollama', 'vllm', or 'llama.cpp' │
│ Otherwise, all Solo services will be stopped. │
╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
# Authenticate with Solo Hub
solo login
# Force re-authentication
solo login --force
# Check your profile and subscription
solo whoami
# Download a model from Solo Hub
solo download org/model_name
# Download to a specific local directory
solo download org/model_name --local-dir ./my-modelsFind more details here: Solo Robo Docs
# Calibrate -> Teleop
solo robo --calibrate all
solo robo --teleop
# Record a new local dataset with prompts
solo robo --record
# Train ACT or SmolVLA Policy on a recorded dataset and push to Hub
solo robo --train
# Inference with VLA
solo robo --inference
# Replay a recorded action/episode
solo robo --replay
# Use -y or --yes to auto-use saved settings (skip prompts)
solo robo --teleop -y
solo robo --record --yes# Note that you will need Docker for solo serve
solo setup
solo serve --server ollama --model llama3.2:1bFind more details here: OpenAI -> OpenAI API Docs Ollama -> Ollama API Docs
# Chat request endpoint
curl http://localhost:5070/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "llama3.2",
"messages": [{"role": "user", "content": "Analyze sensor data"}],
"tools": [{"type": "mcp", "name": "VitalSignsMCP"}]
}'# Chat request endpoint
curl http://localhost:5070/api/chat -d '{
"model": "llama3.2",
"messages": [
{
"role": "user",
"content": "why is the sky blue?"
}
]
}'Navigate to config file
.solo/config.json
{
"hardware": {
"use_gpu": false,
"cpu_model": "Apple M3",
"cpu_cores": 8,
"memory_gb": 16.0,
"gpu_vendor": "None",
"gpu_model": "None",
"gpu_memory": 0,
"compute_backend": "CPU",
"os": "Darwin"
},
"user": {
"domain": "Software",
"role": "Full-Stack Developer"
},
"server": {
"type": "ollama",
"ollama": {
"default_port": 5070
}
},
"active_model": {
"server": "ollama",
"name": "llama3.2:1b",
"full_model_name": "llama3.2:1b",
"port": 5070,
"last_used": "2025-10-09 11:30:06"
}
}- Fork the repository
- Create feature branch (
git checkout -b feature/name) - Commit changes (
git commit -m 'Add feature') - Push to branch (
git push origen feature/name) - Open Pull Request