# Sigma Studio

A modular desktop AI workspace for running, chatting with, fine-tuning and quantizing local models, and orchestrating agents and MCP tools, across GPUs, CPUs and edge devices.

From one interface, Sigma Studio can search, download and organize open-source models from Hugging Face or GGUF repositories with resumable multi-stream downloads, chat with local models through its native SigmaEngine or with cloud providers with low-latency streaming, and switch between predefined role presets such as software architect, coder or security auditor.

It benchmarks hardware (tokens per second, time-to-first-token, VRAM saturation), fine-tunes small language models locally with Unsloth QLoRA and PEFT, and quantizes raw weights into GGUF formats such as Q4_K_M or Q8_0 to match available VRAM.

It connects to Model Context Protocol servers for web browsing, terminal execution and calendar or email access inside a sandbox, and runs entirely locally without a subscription. It is written in Python.

Website: https://github.com/Sigmanih/SigmaStudio