Summary

• Telegram bot now receives voice notes, audio files, and video notes (previously silently dropped)
• Transcribes audio to text via OpenRouter chat completions API (configurable STT provider/model)
• Original audio files preserved as attachments alongside transcribed text
• Graceful fallback: if STT is not configured or fails, audio is still saved as attachment with placeholder text

Configuration

{
  "sttProvider": "openrouter",
  "sttModel": "google/gemini-2.5-flash"
}

Requires OPENROUTER_API_KEY environment variable (or other provider key).

Changed files

• src/config.ts — sttProvider / sttModel fields in AgentConfig
• src/adapters/telegram/transcribe.ts — New transcription module
• src/adapters/telegram/bot.ts — Voice/audio/video_note handling + transcription integration
• src/main.ts — Wires STT config to TelegramBot
• Tests: 16 new test cases across 2 test files

Test plan

• Set sttProvider/sttModel in settings.json and send a voice note to the bot
• Verify transcribed text appears in bot response and log.jsonl
• Verify original .ogg file saved in attachments/
• Test without STT config — voice should save as attachment with placeholder text
• npm test passes (162 tests)

🤖 Generated with Claude Code