# NobodyWho

NobodyWho é um mecanismo de inferência para executar LLMs localmente e com eficiência em dispositivos compatíveis. É voltado a desenvolvedores que buscam inferência de modelos offline, entradas multimodais e fluxos de trabalho locais de voz e texto sem chaves de API.

- Findings24: <https://findings24.com/pt/tools/nobodywho>
- Site: <https://nobodywho.ai>
- Preços: Freemium
- Categorias: Assistente de Código IA

## O que é NobodyWho?

NobodyWho is an on-device inference engine for running large language models locally. The project describes itself as a way to use LLMs efficiently on any device, with an emphasis on offline operation and local execution rather than hosted API calls.

The repository README highlights support for chat models, structured tool calling, multimodal input, text-to-speech, speech-to-text, and model downloading from Hugging Face or any direct URL. It also points to GPU-accelerated inference through Vulkan or Metal, conversation-aware context shifting, and compatibility with GGUF models powered by llama.cpp.

## O que NobodyWho pode fazer?

### Local, offline inference

Runs LLMs without API keys or hidden usage fees, which makes it suitable for private or network-limited workflows.

### Chat model support

Works with chat LLMs such as Gemma, Qwen, and Mistral, with the broader claim of compatibility with many GGUF-format models.

### Type-safe tool calling

Generates structured grammars from function signatures, reducing the need to write schemas manually.

### Multimodal inputs

Accepts image and audio information as part of LLM interactions.

### Speech workflows

Includes text-to-speech backends such as Kokoro, Pocket TTS, and Supertonic, plus speech-to-text via Whisper.

### Flexible model loading

Can load models directly from Hugging Face or from any URL, which simplifies distribution and testing.

## Casos de uso

### Offline assistants and chat apps

Build local chat experiences that do not depend on external model APIs.

### Multiplatform embedded AI features

Add LLM capabilities to Kotlin, Swift, Python, Flutter, React Native, or Godot projects using the documented platform packages.

### Apps that need speech input and output

Combine Whisper transcription with local text-to-speech for voice-enabled workflows.

### Tool-using agents

Use type-safe function calling when you want the model to trigger structured actions from application code.

### Prototype-to-product model testing

Load GGUF models from Hugging Face URLs and experiment with local inference across supported runtimes.

## Perguntas frequentes

### Does NobodyWho require an API key?

No. The README states that it runs locally and offline, with no API keys needed.

### What kinds of models does it support?

The repository says it can run chat LLMs such as Gemma, Qwen, and Mistral, and that it is compatible with thousands of pretrained LLMs in GGUF format.

### Can it handle voice workflows?

Yes. The README mentions text-to-speech and speech-to-text support, including Kokoro, Pocket TTS, Supertonic, and Whisper.

### Which platforms are supported?

The project lists Kotlin, Swift, Python, Flutter, React Native, and Godot.

### Where do models come from?

The README says models can be loaded from Hugging Face or any URL.

## Informações rápidas

- **Category**: Developer Tool
- **Primary purpose**: Local LLM inference
- **Deployment model**: Runs locally/offline
- **Supported platforms**: Kotlin, Swift, Python, Flutter, React Native, Godot
- **Model format**: GGUF
- **Project source**: github.com