Download and run AI models (LLMs) directly in your browser, with no server. Pick a task in the tabs, choose the model and click Download and load model; the first time it downloads from the Hugging Face Hub and is cached (IndexedDB). In chat, the conversation history is included in every inference: the more context, the longer the processing time.
Each tab is a different task: chat / generation, summarize, translate, sentiment analysis, question answering, entity recognition (NER) and zero-shot classification. Each task offers its own models and one-click examples.
It uses Transformers.js + ONNX Runtime Web to run the models in the browser via WebAssembly or WebGPU. None of the text you enter is sent to any server. Speed depends on your hardware: small models (<100 MB) are fast even on CPU; large ones need a dedicated GPU or several minutes.
Completes or generates text from a prompt. Similar to using a local LLM (Ollama style).