ChaosEngineAI
A desktop control plane for discovering, converting, running, chatting with and benchmarking open-weight language and image models entirely on your own machine.
ChaosEngineAI is a desktop application that pairs a Tauri and React interface with a Python backend to drive local model runtimes, including llama.cpp, Apple MLX and optionally vLLM. From one window it can discover and download Hugging Face models, convert them to MLX, load them into a warm pool, serve them over an OpenAI-compatible API, chat with them, benchmark them, and generate images and video through local Stable Diffusion and diffusion video models.
It also supports pluggable KV cache compression strategies, speculative decoding methods that can accelerate generation, and per-chat runtime profiles that remember the exact model, quantization and settings used in a session. The project describes itself as under active development, with breaking changes possible between releases.