Get it running
Run Pravrudhi on your hardware, then pick a target, a benchmark and a budget.
What it needs
For training, use an NVIDIA GPU with at least 8 GB of GPU memory. An Apple Silicon Mac can serve open models. The size of the model you can run depends on the memory available.
Install Docker and keep it running to build the execution image. Use uv to manage Python and install the project dependencies. You will also need Git and make available in your terminal.
Quickstart
Run these commands in order in your terminal. They install the project, prepare your workspace, build the execution image and check your machine before opening the console.
git clone https://github.com/AxisMeru/pravrudhi.git
cd pravrudhi
uv sync
uv run pravrudhi init --root .
make exec-image
uv run pravrudhi doctor
uv run pravrudhi appReview the doctor check and resolve any reported setup issues before continuing. The final command opens this interface at localhost:8008. Keep the terminal running while you use it.
Connect this console to your own engine
The hosted console can point at an engine on your own machine. Set NEXT_PUBLIC_API_BASE in the frontend deployment environment to your engine's browser-accessible URL, then rebuild and redeploy the console. This setting is included at build time.
NEXT_PUBLIC_API_BASE=https://engine.example.comReplace the example with your engine's address. For an HTTPS hosted console, use an HTTPS engine URL that your browser can reach.
The engine only answers the origins its operator names in PRAVRUDHI_ALLOWED_ORIGINS. On the engine machine, allow your console's exact origin, including its scheme and port if present, before starting the engine. Replace the example below with your hosted console's origin.
PRAVRUDHI_ALLOWED_ORIGINS=https://console.example.com uv run pravrudhi appChanging the frontend address does not grant permission on the engine. For local use, open localhost:8008 directly after the quickstart.