Telegram chatbot that answers text and voice messages

Telegram chatbot that answers text and voice messages

An n8n workflow that answers Telegram messages with an AI agent on OpenAI's gpt-5.6-terra. A text goes straight to the agent, and a voice note is turned into text first.

What is inside

• workflow.json: the workflow as n8n exports it, sixteen nodes. Thirteen do the work and three are sticky notes that label the steps. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The chat model was gpt-4o. OpenAI's deprecations page, read on 12 November 2026, shuts the dated gpt-4o snapshots down on 14 January 2027, with gpt-6-luna as the replacement. Mindket's copy uses gpt-5.6-terra. OpenAI's model page lists it with Chat Completions and function calling, so you can add tools later.
• The temperature of 0.7 and the frequency penalty of 0.2 were dropped. The model page lists no temperature setting.
• The step that turned voice into text was n8n's OpenAI node, which always sends whisper-1. OpenAI's page lists whisper-1 for shutdown on 26 February 2027, with gpt-transcribe or gpt-live-transcribe as replacements. Mindket's copy calls OpenAI's transcription address with an HTTP Request node and the model gpt-transcribe.
• OpenAI's speech to text guide lists the file formats ogg and webm among others, but not oga, the ending Telegram gives voice notes. A new Code node, Name the voice file, renames the file to voice.ogg before it goes out.
• Several nodes were linked to credentials in the author's n8n. The links are gone, so select your own.
• Nothing else. The system message, the switch and the closing thank-you line are the file's own.

Before you start

• An n8n instance with a public HTTPS address, on n8n Cloud or your own server. Telegram sends every message to that address.
• A Telegram bot made with BotFather, and its token.
• An OpenAI API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Telegram credential with your bot token. Select it in Listen for incoming events, Send Typing action, Download voice file, Send final reply, Correct errors and Send error message.
3. Add an OpenAI credential and select it in OpenAI Chat Model and Convert audio to text.
4. Check the model in OpenAI Chat Model. The file uses gpt-5.6-terra.
5. Activate the workflow, then send your bot a text and a voice note.

How a run goes

1. Listen for incoming events starts the run for each message. It also starts Send Typing action, so the chat shows a typing sign.
2. Determine content type sorts the message. Text goes on, a voice note goes to Download voice file, and anything else goes to Send error message.
3. A voice note passes through Name the voice file and Convert audio to text, which posts the file to OpenAI's transcription address.
4. Combine content and set properties puts the typed text or the transcript into the field CombinedMessage. It also records whether the message was text or voice and whether it was forwarded.
5. AI Agent reads CombinedMessage. It uses OpenAI Chat Model and keeps the last ten messages of the chat in Window Buffer Memory.
6. Send final reply sends the answer in Telegram's HTML mode, followed by a thank-you line. If Telegram rejects the HTML, Correct errors sends the text with its special characters escaped.

Change the result

• Rules and tone: the system message in AI Agent.
• Model: model in OpenAI Chat Model.
• Memory: contextWindowLength in Window Buffer Memory.
• Closing line: the text in Send final reply.
• Text for unsupported messages: the text in Send error message.

Good to know

• Mindket did not run this workflow. Telegram and OpenAI were not called, and nobody tried the new transcription step with a real voice note. The field names and limits come from OpenAI's and Telegram's docs and from n8n's source, read on 12 November 2026.
• The transcription call posts the multipart fields file and model, as OpenAI's speech to text guide shows. The answer holds the transcript in text.
• Telegram bots can download files up to 20 MB (Telegram's Bot API page). OpenAI's guide allows files up to 25 MB.
• The agent node is version 1.1, which n8n sets to the conversational type by default. Its prompt asks the model for a JSON action.
• A message that starts with a slash still counts as text, because the first rule only checks that the text is not empty.
• Telegram rejects a message longer than 4096 characters.
• Your messages and voice notes go to OpenAI.
• The file holds no credential and no key.