Step 2: Map and Route Incoming Message Types

This step normalizes Telegram payloads and routes the execution based on whether the user sends text, a document, voice note, or a callback.

1. In Map Incoming Context, add fields using these expressions:
   chat_id = {{ $json.callback_query?.message.chat.id || $json.message?.chat.id }}
   callback_data = {{ $json.callback_query?.data || 'none' }}
   message_text = {{ $json.message?.text || 'none' }}
   update_type = {{ (() => { if ($json.message?.text) { return 'text'; } else if ($json.callback_query?.data) { return 'callback'; } else if ($json.message?.document) { return 'document'; } else if ($json.message?.voice) { return 'voice' } else { return 'unknown'; } })() }}
   file_id = {{ $json.message.document?.file_id }}
   file_mime_type = {{ $json.message.document?.mime_type }}
   file_caption = {{ $json.message.caption }}
   voice_file_id = {{ $json.message.voice.file_id }}
2. In Route by Input Type, keep the four rules based on {{ $json.update_type }} to send items to text, document, voice, or callback outputs.
3. Verify the flow: Telegram Intake Trigger → Map Incoming Context → Route by Input Type.

Step 3: Configure Document Handling and Summarization

When a document is sent, the workflow downloads it, converts binary to data, extracts company information, then prepares a prompt for the agent.

1. In Fetch Document File, set Resource to file and File ID to {{ $json.file_id }}.
2. Credential Required: Connect your telegramApi credentials in Fetch Document File.
3. In Encode Document Data, set Operation to binaryToPropery.
4. In Summarize Document Content, set URL to https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent and JSON Body to the provided template using {{ $('Fetch Document File').item.binary.data.mimeType }} and {{ $json.data }}.
5. Credential Required: Connect your googlePalmApi credentials in Summarize Document Content.
6. In Compose Document Prompt, set chatInput to {{ $('Map Incoming Context').item.json.file_caption }} --- Information extracted from document sent by the user --- {{ $json.candidates[0].content.parts[0].text }} --- End for information extracted from document sent by the user ---.

Step 4: Configure Voice Note Transcription

Voice notes are fetched from Telegram, converted to inline audio data, transcribed with Gemini, and then formatted for the agent.

1. In Fetch Audio File, set Resource to file and File ID to {{ $json.voice_file_id }}.
2. Credential Required: Connect your telegramApi credentials in Fetch Audio File.
3. In Encode Audio Data, set Operation to binaryToPropery.
4. In Transcribe Audio Content, keep URL as https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent and ensure the JSON body includes {{ $json.data }} inside inline_data.
5. Credential Required: Connect your googlePalmApi credentials in Transcribe Audio Content.
6. In Compose Audio Prompt, set chatInput to {{ $json.candidates[0].content.parts[0].text }}.

Step 5: Set Up the Social Content Agent and AI Tools

The agent orchestrates content drafting and publishing by leveraging Gemini as the LLM plus multiple AI tools for posting, knowledge retrieval, and memory.

1. Open Social Content Agent and verify the System Message matches the content manager instruction block in the node.
2. Connect Gemini Chat Engine as the language model for Social Content Agent. Credential Required: Connect your googlePalmApi credentials in Gemini Chat Engine (configured via the agent’s language model connection).
3. Connect Gemini Embedding Builder to Lookup Knowledge Records. Credential Required: Connect your googlePalmApi credentials in Gemini Embedding Builder (configured via the agent’s tool connections).
4. Ensure Short-Term Memory Buffer uses Session Key {{ $('Map Incoming Context').item.json.chat_id }} with Session ID Type set to customKey, and is connected to Social Content Agent for memory.
5. Confirm the AI tools are connected to Social Content Agent: Publish X Post, Publish LinkedIn Company Post, Publish LinkedIn Personal Post, Lookup Knowledge Records, and Store Knowledge Tool.
6. Credential Required: Connect your twitterOAuth2Api credentials for Publish X Post and linkedInOAuth2Api credentials for both Publish LinkedIn Company Post and Publish LinkedIn Personal Post (managed via the agent’s connected tools).

Step 6: Configure Callback Memory Clearing and Telegram Responses

This step handles callback cleanup, splits long responses, and sends the final message back to Telegram.

1. In Callback Condition Check, keep the condition callback_data equals cm using {{ $('Map Incoming Context').item.json.callback_data }}.
2. Confirm the flow: Callback Condition Check → Purge Conversation Memory → Confirm Memory Cleared.
3. In Send Callback Error, keep Text as Invalid callback and Chat ID as {{ $('Map Incoming Context').item.json.chat_id }}.
4. In Segment Agent Output, keep the JavaScript splitter to chunk item.json.output into 3000-character parts.
5. In Expand Message Parts, set Field to Split Out to parts.
6. In Dispatch Telegram Reply, set Text to {{ $json.parts || "Error occured, pls try again" }} and Chat ID to {{ $('Map Incoming Context').item.json.chat_id }}.
7. Credential Required: Connect your telegramApi credentials in Dispatch Telegram Reply, Send Callback Error, and Confirm Memory Cleared.

Step 7: Test & Activate Your Workflow

Run end-to-end tests for text, document, voice, and callback inputs before enabling the workflow in production.

1. Click Execute Workflow and send a Telegram text message to confirm Prepare Text Prompt → Social Content Agent → Dispatch Telegram Reply works.
2. Send a document to verify the flow Fetch Document File → Encode Document Data → Summarize Document Content → Compose Document Prompt → Social Content Agent.
3. Send a voice note to verify Fetch Audio File → Encode Audio Data → Transcribe Audio Content → Compose Audio Prompt → Social Content Agent.
4. Test the callback by sending a cm callback to confirm Purge Conversation Memory and a response from Confirm Memory Cleared.
5. When successful, toggle the workflow to Active for live use.