Telegram chat bot that answers from a Google Sheet schedule

Telegram chat bot that answers from a Google Sheet schedule

An n8n workflow that turns a Telegram bot into an assistant for a schedule kept in a Google Sheet. A member writes a
question to the bot, and an AI model answers from the current rows of the sheet. A second door, n8n's own chat, lets
you test the same flow without Telegram.

What is inside

• workflow.json: the workflow as n8n exports it, 15 nodes and 8 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Settings held the original author's sheet address. It now reads <YOUR_GOOGLE_SHEET_URL>. The Schedule node
  also carried that sheet's address and tab name as saved labels. Both are gone, and the node still reads the first tab,
  gid=0.
• The LLM node, an OpenRouter Chat Model, named no model, so it ran n8n's default for that node, openai/gpt-4.1-mini.
  OpenAI shuts that model down on 14 January 2027 and names gpt-6-luna as the replacement. The node now names
  openai/gpt-6-luna outright, which OpenRouter's public model list carries. This agent has no tools, so gpt-6-luna
  fits.
• SetResponse held a second, empty field with no name. It is removed.

Before you start

• An n8n instance with a public HTTPS address, such as n8n Cloud, if you want the Telegram door. The chat in the editor
  needs no address.
• A Telegram bot token. You get one by writing to BotFather inside Telegram.
• A Google account that can open your sheet, and a Google Sheets OAuth2 credential in n8n.
• An OpenRouter API key.
• A sheet with a header row and one event per row on its first tab.

Set it up

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Add a Telegram credential with your bot token to telegramInput, SendTyping and telegramResponse.
3. In Settings, replace <YOUR_GOOGLE_SHEET_URL> with the address of your sheet as your browser shows it. Add a Google
   Sheets OAuth2 credential to Schedule. The node reads the first tab. To read another tab, pick it in Schedule.
4. Add an OpenRouter credential to LLM.
5. Open ScheduleBot and rewrite its system message for your own group. It now talks about a meetup group.
6. To test, open the chat at the bottom of the canvas and ask a question about your schedule.
7. For Telegram, save the workflow and switch it on, then write to your bot.

How a run goes

1. A run starts at telegramInput, the Telegram Trigger, or at n8nInput, the chat in n8n. The Telegram Trigger listens
   for message updates only.
2. On the Telegram side, SendTyping shows the typing sign in the chat, and telegramChatSettings copies the message
   text and the chat id into inputMessage and chatId, with mode set to telegram. On the n8n side,
   n8nChatSettings copies the chat text and the session id into the same fields, with mode set to n8n.
3. Settings adds scheduleURL. Schedule reads the sheet. ScheduleToMarkdown turns every row into one Markdown
   table in markdown, with the first row's column names as the header.
4. ScheduleBot is an agent on LLM, with Memory beside it. Its system message holds the table, and the member's
   text is the prompt. It has no tools.
5. SetResponse copies the answer into responseMessage. Switch then follows mode. For Telegram, telegramResponse
   sends the answer to the same chat. For n8n, n8nResponse is a No Operation node that ends the run, and the chat
   shows the answer.

Change the look

• The schedule: scheduleURL in Settings, and the tab in Schedule.
• Who the bot is and how it answers: the system message in ScheduleBot.
• The model: LLM.
• How much it remembers: the window length in Memory. It keeps one conversation per chat id.
• The reply format: telegramResponse has the text and the options of a Telegram send.

Good to know

• The facts below come from n8n source, Telegram's Bot API page and OpenRouter's public model list, read on 12
  November 2026. Nothing was run: no bot, no sheet and no model call.
• Telegram sends updates to an HTTPS address, so an n8n that is not reachable that way cannot receive bot messages. The
  chat in the editor works anywhere.
• Anyone who finds your bot can ask it, and the answers come from your sheet. telegramInput is version 1.1. That version
  shows the Restrict to Chat IDs and Restrict to User IDs fields but does not apply them. n8n filters only from version
  1.2, so update the node before you rely on them.
• The whole table goes into the system message of every question, and the sheet is read again each time. A long schedule
  costs more per message and slows each answer.
• Schedule uses Get Row(s), which adds a row_number column to every row. The model sees it as the first column of
  the table.
• ScheduleToMarkdown takes its column names from the first row. A cell that holds a vertical bar can break the table.
• Memory keeps the last five messages by default, one conversation per chat id, in n8n's own process. A restart of n8n
  clears it.
• telegramResponse sends with Telegram's version 1.2 defaults: Markdown parse mode, link previews off, and the line
  "This message was sent automatically with n8n" added under every answer. An answer with a stray _ or * can make
  Telegram refuse the message.
• SendTyping shows the typing sign for 5 seconds or less, and Telegram clears it when the answer arrives.
• Billing: OpenRouter bills each answer. Its public list shows openai/gpt-6-luna at USD 0.1 per million tokens in and
  USD 0.5 per million out. The model's reasoning tokens count as output.
• Your data: the sheet's rows and each message go to OpenRouter and the model's provider. The answers and the typing
  sign go through Telegram.
• Credentials: the JSON holds none.