CloverPal
An open-source local LLM client for Android that runs GGUF models and character group chats offline, with on-device speech recognition.
CloverPal is an Android app for people who want private AI chat and character roleplay on their phone. Its Offline Silly Tavern mode lets you create characters with their own avatars and system prompts, then have them converse in group chats. Model inference runs on-device, and chats and uploaded files stay on your phone.
The app uses llama.cpp to run GGUF models, including Qwen3.5, Qwen3, Gemma 4, Gemma 3 and LLaMA. You can import model files you already have or download them from HuggingFace and ModelScope. Downloads need internet access; chat works offline once the model is on your device. For supported models, you can turn thinking on or off and limit how much generation it uses.
Beyond roleplay, CloverPal can read TXT, DOCX and PDF files into a conversation's context. It renders mathematical notation with KaTeX and displays HTML code previews inside the chat. Offline speech recognition uses whisper.cpp and detects the spoken language automatically, so voice input doesn't depend on a cloud transcription service.
You can edit messages, regenerate answers, search chat history and star conversations. The interface supports Chinese, English, Japanese, Korean, Spanish and French, with adjustable colors, backgrounds and text size. CloverPal is open source under AGPL-3.0 and runs on Android 9 or later.
