MoFa

A mobile-first LLM chat client for Android.

English | 简体中文

Android APK & PWA · Bring your own API key · 100% local data

MoFa is an open-source AI chat app for Android. Connect your own API keys (OpenAI-compatible, Claude, Gemini, DeepSeek, SiliconFlow, OpenRouter, Ollama…) and chat with streaming responses, markdown, code highlighting, image input and reasoning models. All conversations and keys stay on your device.

Features

• Multi-provider — OpenAI-compatible APIs, Claude (Anthropic) and Gemini native protocols; per-provider API key & base URL
• Streaming chat — real-time token streaming with markdown rendering and code syntax highlighting
• Thinking mode — show model reasoning (DeepSeek-R1, QwQ…), adjustable thinking effort (low / medium / high)
• Image input — send photos to vision models, auto-compressed
• Per-chat model — each conversation remembers its own provider & model
• Web search — optional free web search (Bing / DuckDuckGo / SearXNG, no API key, scraped client-side); the model decides whether a search is needed, formulates the query and cites numbered sources
• File I/O — attach documents (PDF / Word / text / code, extracted client-side) and save replies as real files (.md / .txt / .json / .docx, share or download)
• Context management — auto-truncate history by rounds or token budget, with a model-window safety net
• Assistant presets — per-conversation system prompt with built-in presets
• Sessions — local history, rename, export to Markdown
• Auto update — in-app download with progress; install permission is requested once, then the update resumes automatically and the downloaded APK is cleaned up

Download

• Android 6.0+: Download MoFa v1.0.1 APK
• iOS / any browser: PWA — deploy dist/ to any static server, then "Add to Home Screen" in Safari

Development

npm install
npm run dev        # dev server, reachable from phones on the same Wi-Fi
npm run mock       # local mock LLM server (OpenAI / Anthropic / Gemini)
npm run build      # production build to dist/
npm run typecheck

Testing

npm run test:smoke   # build + 16 regression checks against a local mock LLM (~60s)

Full checklist (manual + native Android) is maintained locally in docs/TESTING.md and not committed to the repo.

Build the Android APK (requires Android Studio):

npm run release 1.0.0 "release notes"
# → releases/mofa-1.0.0.apk + releases/latest.json (for in-app updates)

Tech Stack

Vue 3 · TypeScript · Vite · Pinia · Dexie (IndexedDB) · Capacitor 7 · markdown-it + highlight.js

Project Structure

src/
├── components/       # UI (chat/, ui/)
├── composables/      # useToast / useDialog / useUpdate / useBackHandler
├── db/               # Dexie: conversations & messages
├── lib/              # markdown pipeline, image compression, model abilities
├── services/
│   ├── llm/          # protocol adapters: openai-compatible / anthropic / gemini
│   ├── update.ts     # in-app update
│   └── stats.ts      # anonymous usage stats (opt-out in settings)
├── stores/           # Pinia: settings / providers / chat
└── views/            # ChatView / SettingsView / ProviderEditView

License

MIT