Docs
Setup

OmniChain is distributed as an open-source project on GitHub. The main branch will always contain the latest stable software version.

OmniChain (Core)

Requirements

  • Windows / Linux / macOS
  • Git
  • NodeJS v20.11.0 or higher

Installation

  1. Clone the repository:
git clone https://github.com/zenoverflow/omnichain
  1. Install dependencies:
cd omnichain
npm install
  1. Start the app:
npm run serve
  1. Open your browser and navigate to http://localhost:12538.

Update

Simply pull the latest changes from the repository and reinstall the dependencies:

git pull
npm install

Server configuration

By default, the main server runs on port 12538, and the OpenAI-compatible API runs on port 5002. If the Python module is installed and running, it is expected on http://localhost:12619.

You can change these defaults by passing the following arguments to the npm run serve command:

npm run serve -- --port 12345 --port_openai 5003 --module_python_url http://localhost:12619

External Python Module (Optional)

To support Python-only functionality and the ability to use custom Python code, OmniChain provides an external Python module that can be installed and run separately from the main app.

Requirements

  • Windows (WSL) / Linux / macOS
  • Git

Requirements (Nvidia GPU)

  • CUDA Drivers & Compatibility with CUDA >= 11.8

Requirements (AMD GPU)

  • ROCm Drivers & Compatibility with ROCm >= 6.1

Gotchas

Windows Subsystem for Linux (WSL)

The installation on WSL will only work if the project is on a path inside your WSL instance, like /home/youruser/PROJECTS. If you try to setup and run on the regular Windows filesystem, you will run into unfixable issues with symlinks.

macOS

CUDA and ROCm are not supported on macOS. Hardware acceleration is up to the devs of Pytorch and the other underlying libraries.

Installation

  1. Clone the repository:
git clone https://github.com/zenoverflow/omnichain_external_python
  1. Make sure the setup script is executable:
chmod +x setup.sh
  1. Run the setup script:
  • For Nvidia GPUs (Linux only; CUDA version can be 11.8 / 12.1 / 12.4):
./setup.sh --device=cuda --cuda_version=12.4
  • For AMD GPUs (Linux only):
./setup.sh --device=rocm
  • For CPU-only
./setup.sh --device=cpu
  • For MacOS
./setup.sh
  1. Make sure the server script is executable:
chmod +x start.sh
  1. Run the server:
  • Default (host: 127.0.0.1; port: 12619):
./start.sh
  • Customise host and port:
./start.sh --host=0.0.0.0 --port=8080

Note: If you change the host/port, or you run the server on a different machine, do not forget to update the module_python_url argument when starting the core app.

Update

Simply pull the latest changes from the repository and rerun the setup script with the same arguments you used during the initial setup:

git pull
./setup.sh --device=cuda --cuda_version=12.4

Get help / Verify options

Both the setup and start scripts support the -h flag to display the available options. This is useful if the documentation happens to be temporarily outdated:

./setup.sh -h
./start.sh -h
Quick-Start Guide