This project provides a production-ready server CLI for integrating specialized AI capabilities into physical systems, targeting developers working with Physical AI and robotics. It streamlines the deployment and management of AI models, offering modular components and FastMCP integration for enhanced inference server functionality.

How It Works

Solo Server acts as a central command-line interface for managing AI inference and robotics operations. It leverages modular Python components to add AI capabilities to existing inference servers. The architecture supports various model serving backends like Ollama, vLLM, and llama.cpp, offering OpenAI-compatible API endpoints for seamless integration. Its core advantage lies in unifying model management, robotics control, and data handling within a single, accessible CLI tool.

Quick Start & Requirements

Installation begins with the uv package manager, available via curl (Mac/Linux) or Powershell (Windows). A Python virtual environment using Python 3.12 is recommended (uv venv --python 3.12). Solo Server can be installed via PyPI (uv pip install solo-server) or from source (git clone ..., uv pip install -e .). An automated installation script (install_mac.sh) is provided for Mac users, handling uv, Python 3.12.12, and development setup. Activation requires sourcing the virtual environment (source .venv/bin/activate). Docker is required for the solo serve command. Fallback handling for mujoco dependencies is noted. A quickstart video is available here.

Highlighted Details

• Comprehensive CLI: Offers commands for setup (solo setup), robotics operations (solo robo - calibration, teleoperation, recording, training, inference), model serving (solo serve), status checks (solo status), model listing/downloading (solo list, solo download), testing (solo test), and stopping services (solo stop).
• Robotics Integration: The solo robo command suite provides direct control for robotics tasks, including interactive Lerobot support for motor calibration, teleoperation, data recording, and training Diffusion Policies.
• Model Serving Flexibility: Supports Ollama, vLLM, and llama.cpp, providing OpenAI and Ollama compatible API endpoints (e.g., http://localhost:5070/v1/chat/completions).
• FastMCP Integration: Designed for integration with FastMCP for advanced Physical AI inference capabilities.

Maintenance & Community

The repository includes a standard "Contributing" guide outlining a typical Git workflow for pull requests. Specific details regarding community channels (like Discord/Slack), active maintainers, sponsorships, or a public roadmap are not detailed in the provided README.

Licensing & Compatibility

The project is licensed under the MIT License, permitting broad use and modification. No specific compatibility notes regarding linking with closed-source applications are mentioned.

Limitations & Caveats

The README indicates default CPU usage (use_gpu: false in config) and notes fallback handling for mujoco dependencies, suggesting potential complexities or manual configuration required for optimal GPU acceleration and specific hardware integrations. The automated Mac installer specifies Python 3.12.12, while the badge indicates Python 3.9+, indicating potential version sensitivity.