NobodyWho
An on-device inference engine for running LLMs locally and efficiently on any device, with bindings for Godot, Kotlin, Swift, React Native, Flutter and Python.
NobodyWho is an inference engine, built on llama.cpp, for running large language models locally and offline, with no API keys or hidden fees. It runs any chat LLM in GGUF format, such as Gemma, Qwen or Mistral, generates type-safe tool calls by turning function signatures directly into structured grammars, and accepts image and audio input for multimodal use.
Beyond text generation, it can synthesize speech with Kokoro, Pocket TTS or Supertonic, transcribe audio with Whisper, and detect voice activity with Silero to know when to start and stop listening. It accelerates inference with Vulkan or Metal, and ships bindings for Godot, Kotlin, Swift, React Native and Expo, Flutter and Python, covering desktop, Android, iOS, visionOS and watchOS; it is written in Rust under the EUPL-1.2 license.