Local LLM Autonomously Draws a Dog in MS Paint via Open WebUI
A user successfully configured a local QFN 3.8 NVFP4 model using Open WebUI 0.13.0 and Open Terminal 0.17.2 to autonomously control a virtual desktop mouse and draw a dog in MS Paint. This capability utilizes new computer use and desktop integration features introduced in these tool versions. This demonstrates a practical, cutting-edge milestone where local AI models can interact directly with graphical user interfaces using mouse coordinates rather than relying solely on generated code or text. It expands the scope of open-source local agents toward generalized desktop automation. The virtual desktop environment ran inside an unactivated Windows 11 ISO customized via an agent, though Open Terminal natively supports sandboxed Linux desktops. The setup remains fully interactive within the browser, allowing human intervention when necessary.
## BACKGROUND
Open WebUI is a self-hosted graphical interface for interacting with large language models completely offline or via remote APIs. NVFP4 quantization is a low-precision optimization technique that maintains floating-point semantics with a shared exponent to improve dynamic range on supported hardware.