Back to TemplatesPersonal assistant bot with multi-agent system using Telegram & Google Gemini
Last update
Last update 3 months ago
Categories
How It Works
- Telegram Trigger receives incoming messages (text, voice, photo, document).
- Switch routes by message type to appropriate processors:
- Text → forwarded as-is.
- Voice → downloaded and sent to Transcribe a recording.
- Photo → downloaded, converted to base64, then sent to Analyze image.
- Document → routed to document handler.
- Merge collects the processed input and passes a unified prompt to Manager Agent.
- Manager Agent (LM: Google Gemini) orchestrates specialized agents/tools:
memory_base (Airtable) → saving & retrieving personal/company memory
todo_and_task_manager (Todoist / Google Sheets) → tasks
email_agent (Gmail) → composing/sending emails
calendar_agent (Google Calendar) → scheduling
research_agent (SerpAPI / Wikipedia / Wolfram) → web research
project_management (Google Sheets) → project updates
- Manager Agent updates memory windows and sends the final reply back to Telegram.
Setup Steps
- Create and configure Telegram bot; set bot token/webhook in Telegram Trigger and Telegram nodes. Update
chatId placeholders.
- Add Google Gemini (PaLM) credentials in the Gemini model nodes.
- Configure Airtable knowledge-base: set base ID & table IDs used by
memory_base nodes.
- Connect Google APIs: Sheets, Calendar, Gmail credentials and set document/sheet IDs.
- Configure Todoist, SerpAPI, WolframAlpha credentials and any other tool API keys.
- Verify Window Buffer Memory
sessionKey values (match user sessions).
- Check schedule triggers (cron expressions) and adjust times/timezone.
- Run quick tests: send text, voice, image, and confirm replies and memory writes.
Estimated Setup Time
- 30–60 minutes → if credentials & IDs are ready.
- 2–4 hours → full setup (API keys, spreadsheets, Airtable, detailed permissions).
- 4–8 hours → complex deployment (team permissions, multiple calendars, advanced tool tuning, production testing).