Kobold.cpp

KoboldCpp packs local AI text generation into a single self contained executable with no installation required. Built on llama.cpp and descended from the KoboldAI project, it runs GGML and GGUF models with broad backward compatibility, then layers on far more than text: image generation with Stable Diffusion, SDXL, SD3, and Flux, speech recognition through Whisper, text-to-speech, and multimodal image understanding. The bundled KoboldAI Lite web UI offers chat, instruct, adventure, and story writing modes with editing tools, while the server exposes KoboldCpp, OpenAI, and Ollama compatible API endpoints plus tool calling and MCP support. Precompiled binaries cover Windows, macOS, and Linux with GPU offloading, and it also runs on Google Colab, Docker, and Android through Termux. Hobbyists running roleplay and story models locally form the core audience, alongside anyone wanting a zero setup local inference server. Released under the AGPL-3.0 license.