airunner

AI Runner is an offline, privacy-first inference engine that combines AI art generation with a customizable chat companion. It runs on your own hardware by default, without requiring an internet connection, API key or subscription. Main capabilities: - AI companion: name, personality, voice and long-term memory built from conversations, with awareness of time, date and local weather. - Layered canvas: draw, paint, generate and filter on multiple layers; sketch-to-image, image-to-image, inpainting, compositing and post-process filters. - Image generation: SDXL and Z-Image Turbo with LoRA, embeddings, background removal and configurable safety filters. - Voice: full text-to-speech and speech-to-text, so you can speak with the companion and hear responses in a chosen voice. - Model management: built-in HuggingFace and Civitai downloaders, support for multiple local LLMs and image models. Deployment options include a desktop download, Docker images (GUI, headless and Linux variants), PyPI packages and a web version. A headless HTTP API server exposes endpoints for health, LLM, art, TTS and STT, plus Ollama-compatible and OpenAI-compatible endpoints for integration with tools such as VS Code. Environment variables and CLI flags control model paths, ports, authentication and service enablement. The project is written in Python 3.13, licensed GPL-3.0, and primarily targets Linux with experimental Windows support. It is developed by Capsize LLC and accepts sponsorship.