CloverPal is a hardcore, fully offline LLM client for Android. Powered by deeply optimized llama.cpp (ARMv8.4 DotProd), featuring HTML Live Preview, file reading, LaTeX rendering, Speech Recognition, and an innovative Offline Silly Tavern mode for multi-character roleplay. 100% on-device, zero data leaves your phone.
For Android (v1.1.6): Download CloverPal Here
For iOS: Coming Soon
- Minimalist UI — Clean interface focused on readability, with full theme customization (accent color, light/dark/custom background, font sizes).
- 7-Language Interface — Simplified Chinese, Traditional Chinese, English, Japanese, Korean, Spanish, and French.
- Document Parsing — Read
.txt,.docx, and.pdffiles directly into chat context. - LaTeX Rendering — Math expressions rendered beautifully via KaTeX.
- HTML Live Preview — Preview HTML code blocks in a sandboxed iframe, right inside the chat.
- Offline Silly Tavern — Create characters with custom avatars and system prompts, then let them talk to each other in group chats (2–3 AI characters).
- Speech Recognition — Offline voice-to-text powered by whisper.cpp, supporting automatic language detection across 99 languages.
- Smart Chat Management — Message editing, response regeneration, starred conversations, batch delete, auto-generated titles, and full chat history search.
- Flexible Model Import — External direct-read (saves space) or internal copy (more stable), plus one-tap download from HuggingFace / ModelScope with curated presets.
- Architecture Support — Qwen3.5, Qwen3 (Soft Thinking), Gemma 4, Gemma 3, LLaMA, and other GGUF-compatible models.
- Think Mode Control — Toggle thinking on/off per model, choose format (Qwen3.5 / Qwen3 / Gemma 4), and set a token budget for the thinking process.
- OS: Android
- SDK: Minimum SDK 28 (Android 9.0), Target SDK 35 (Android 15)
- Inference Engine: llama.cpp
- Speech Engine: whisper.cpp
This application provides a framework for local computation only. CloverPal operates strictly offline and does not collect, store, or transmit any private user data, chat histories, or uploaded files to any external servers. The text generated by the AI models loaded into this application does not reflect the views or positions of the developer. Users are solely responsible for the models they choose to import and the outputs generated during usage.
This project is licensed under the GNU AGPL-3.0 License — see the LICENSE file for details. (You are free to use, modify, and distribute this software, but any modified versions or services utilizing this code must also be open-sourced under the same AGPL-3.0 license.)
CloverPal is built using the amazing work from:
- llama.cpp — Efficient on-device LLM inference.
- whisper.cpp — Offline speech-to-text.
- Capacitor — Cross-platform native runtime.
- marked — Markdown rendering.
- mammoth.js — DOCX text extraction.
- KaTeX — LaTeX rendering.
- pdf.js — PDF reading.
And many other open-source libraries that make this project possible!
If you find CloverPal useful, please consider giving this repo a star ⭐ — it's free, takes just one second, and means a lot to an independent developer! Every star helps more people discover CloverPal and motivates continued development. Thank you for your support! 🍀