A desktop GUI for Ollama that knows your hardware, recommends the right model, and keeps everything private — no cloud, no subscriptions.
dataclasses and json:LocalDesk wraps Ollama in a polished desktop experience designed for developers who want full control.
Scans your CPU, RAM, and NVIDIA VRAM on startup and recommends the largest model that will run comfortably — no guessing.
Responses stream token-by-token with a blinking cursor. Incremental UTF-8 decoding means nothing breaks on non-ASCII output.
Keeps context across the last 12 messages so the model actually remembers what you were just talking about.
Drop a .py or .txt file into the chat and it's included inline. Vision models get images passed directly.
Every conversation can be saved as JSON and reopened later. Your chat history stays local, always.
Kill a running response instantly. The partial output is kept so you can see what the model had before you stopped it.
The core is solid. These are next.
Works on Linux, Windows, and macOS. Requires Ollama to be installed and available in your PATH.
⚠️ Windows Defender may warn — no digital signature yet. You can always run from source.