Cross-platform Telegram client built with Tauri 2.0 + React 19 + TypeScript. Features voice transcription, media handling, and optimized message loading.

The same UI also runs in the browser against a remote vasya-server (the Telegram session host) — no Tauri runtime required. Self-hosted: bring your own server and domain.

• Connect: open the web app, enter your vasya-server URL, then sign in with email + password (issues a JWT) or paste an access token. Builds can pre-fill the API origin via the VITE_VASYA_API_URL build var.
• Build: VITE_VASYA_API_URL=https://your-api.example.com npm run build → serve dist/.
• The engine transport is abstracted (src/transport/): Tauri IPC on desktop, HTTP/SSE in the browser. The production stack (vasya-server + sync backend + Caddy) lives in backend/deploy/ — see its README to deploy your own.

The engine (vasya-core) is fronted by vasya-server — an API gateway you can drive programmatically or hand to AI agents:

• REST (/api/v1, ~46 routes) + GraphQL (queries, mutations, WS subscriptions) + SSE event stream (/api/v1/events) + an OpenAPI spec.
• Agent-native: scoped API keys (vk_…) with a per-key permission scope, an audit log, and idempotency-key replay — safe to give an autonomous agent.
• MCP server (vasya-mcp/): wraps the REST API as 29 Model Context Protocol tools at REST parity, so an LLM can run the login flow, manage accounts, create/delete chats, send & download media, edit folders/tabs, search and read/send messages — the full messenger. Configure with VASYA_API_URL + VASYA_AGENT_KEY.

Same UI, four ways to run the engine — pick per use case:

1. Embedded (default desktop) — the Telegram engine runs in-process via Tauri IPC. No server, no backend.
2. Embedded + local API — the desktop app mounts the API server on 127.0.0.1 (Settings → General → Local API server, token-auth). Lets local AI agents / MCP talk to your running app — no separate backend needed.
3. Remote / web — the browser UI talks to a remote vasya-server.
4. Remote / desktop — the desktop app points at a remote vasya-server instead of its embedded engine.

Run vasya-server standalone or alongside the sync backend (backend/deploy/). Everything is self-hostable — bring your own server, domain, and secrets.

• Messaging UX: Telegram-style date separators, rich link previews, and message grouping

• Telegram Integration: Full MTProto support via grammers-client v0.8

• Voice Transcription (STT):

  • Cloud: Deepgram Nova-2 — bring your own API key (Settings → STT); no key is embedded in the build
  • Local: Whisper.cpp via sidecar (~1 GB RAM)
  • Auto-transcribe incoming voice messages
  • Multi-language support (Russian, English, Ukrainian, German, French, Spanish)

• Media Handling:

  • Auto-download for photos/stickers/voice — viewport-scoped (only media actually on screen is fetched)
  • Click-to-download for videos/documents (user downloads jump the queue)
  • Media caching and queue management
  • Paste images directly (Ctrl+V)

• Performance:

  • Message virtualization with @tanstack/react-virtual
  • Optimistic updates with Zustand 5
  • Connection state monitoring

• Multi-account: Switch between multiple Telegram accounts

Frontend:

• React 19 + TypeScript
• Zustand 5 (state management)
• @tanstack/react-virtual (virtualization)
• Vite (build tool)

Backend:

• Tauri 2.0 (Rust + system APIs)
• grammers-client 0.8 (Telegram MTProto)
• whisper-rs + whisper.cpp (local STT)
• reqwest (HTTP client for Deepgram)

• Node.js 20+
• Rust (latest stable)
• Xcode Command Line Tools (macOS)

# Install dependencies
npm install

# Create .env file (Deepgram key is NOT a build input — users add it in Settings → STT)
echo "TELEGRAM_API_ID=your_api_id" >> .env
echo "TELEGRAM_API_HASH=your_api_hash" >> .env

# Run in development mode
npm run tauri dev

# Build for production
npm run tauri build

cd src-tauri/stt-sidecar
cargo build --release
cp target/release/stt-sidecar target/release/stt-sidecar-aarch64-apple-darwin

1. Add repository secrets:
   • TELEGRAM_API_ID
   • TELEGRAM_API_HASH
2. Push to main or trigger workflow manually
3. Download APK from Artifacts (~15-30 minutes)

See .github/ANDROID_BUILD.md for details.

# Install Android Studio + SDK + NDK r25c
export ANDROID_HOME=$HOME/Library/Android/sdk
export NDK_HOME=$ANDROID_HOME/ndk/25.0.8775105

# Add Rust targets
rustup target add aarch64-linux-android armv7-linux-androideabi

# Build
npm run tauri android init
npm run tauri android build

The Deepgram API key is a runtime user setting (Settings → STT), not a build input — it is never embedded in the binary.

MIT

Евгений (suenot)

• GitHub: @suenot