Turn your home computer into a private AI server with a single command
Enter one line in the terminal to run an AI server capable of chat, voice, and image generation directly on your computer. ODS is a deployment system that automatically installs and connects the various software required to run local AI. It is designed so that all data and processing occur within your computer without using the cloud.
Status of the V3 public test phase
ODS is currently undergoing public testing and improvements ahead of the official V3 release. You can use the fixed source snapshot v3.0.0, and it is recommended to pin tags or audited commits for reproducibility. Since full fleet qualification is not yet complete, you should review known operational limitations and remaining release gates. The September update follows the development main branch, which is not a new stable version, so caution is required.
AI features bundled into a single stack
ODS provides pre-connected services for everything from model inference to voice processing and workflow automation.
- Chat and inference Open WebUI supports conversation history, web search, and document uploads. llama-server automatically selects a model suitable for your GPU to handle inference.
- Voice and agents Whisper and Kokoro handle speech recognition and synthesis. Agents like Portal and Hermes call tools and perform tasks, while n8n automates workflows.
- Knowledge retrieval and generation Qdrant vector database and SearXNG perform local document search and web search. ComfyUI is used to generate images locally.
Automatic model selection based on hardware
The installer detects your GPU and memory to automatically select the most suitable model. Tiered model mappings are provided for various hardware, including NVIDIA, AMD Strix Halo, Apple Silicon, and Intel Arc. For example, an NVIDIA card with 24GB of VRAM defaults to the Qwen3.5 27B model, while an AMD system with 64GB of unified memory defaults to the Qwen3.6 35B-A3B model. Using bootstrap mode allows you to download a smaller model first and start chatting within two minutes, while the full model downloads in the background.
Pre-installation checks and limitations
On Linux and macOS, Docker must be installed and running, while Windows requires WSL2 and Docker Desktop. Windows installation requires at least 40GB of disk space and enabled hardware virtualization. Windows users with NVIDIA GPUs must update their drivers to version 570 or higher. Since V3 is currently in the public test phase, you must check the release notes and verification status before using it for critical purposes. Existing native Windows installation environments are not automatically migrated, so caution is needed before switching.
