Build a Voice RAG Chatbot with ElevenLabs and OpenAI in n8n

What This Workflow Does

This workflow listens to spoken questions from customers using an ElevenLabs voice agent.
It finds the best info from stored documents using OpenAI and Qdrant vector search.
Then, it answers back in a natural voice using text-to-speech.
This helps businesses answer questions faster with a human-like voice assistant.

It saves time by automating repetitive calls and gives accurate replies based on current documents.

Who Should Use This Workflow

This workflow fits small business owners who want voice AI answering customer questions.
It is good for those with documents stored in Google Drive who want quick setup without deep coding.

Users should have basic knowledge of API keys and webhook URLs.
It works best if conversation memory is needed to keep answers relevant in longer talks.

Tools and Services Used

• ElevenLabs Voice Agent: Captures spoken questions and plays voice responses.
• OpenAI API: Generates text answers and creates embeddings for document vectors.
• Qdrant Vector Database: Stores document chunks as vectors and retrieves similar content.
• Google Drive: Hosts documents to be indexed and used as knowledge source.
• n8n Automation Platform: Connects and orchestrates all components with nodes and workflows.
• LangChain Nodes: Handle document splitting, embeddings, and memory for context.

Inputs, Processing, and Outputs

Inputs

• Voice questions from customers captured by ElevenLabs agent via webhook.
• Text documents stored in Google Drive folder.

Processing Steps

• ElevenLabs sends question text to n8n webhook.
• Workflow converts documents from Google Drive into smaller chunks.
• OpenAI creates embedding vectors from these chunks.
• Qdrant stores embeddings and searches them when questions come in.
• AI Agent node retrieves top relevant info and generates a conversational answer.
• Conversation memory keeps context across multiple user messages.
• Answer text is sent back to ElevenLabs for voice response.

Outputs

• Natural-sounding voice reply to the customer’s question.
• Updated conversation context for smoother dialogue.

Beginner Step-by-Step: How to Use This Workflow in n8n

1. Importing the Workflow

1. Download the workflow file using the Download button on this page.
2. Open the n8n editor you have access to.
3. Use the Import from File option in n8n to load the workflow.

2. Configuring Credentials and IDs

1. Add your API Keys for OpenAI, ElevenLabs, and Qdrant under n8n credentials.
2. Update any IDs like Google Drive folder ID or Qdrant collection names if needed.
3. Check webhook URLs from ElevenLabs are correctly entered into the Webhook node.

3. Testing the Workflow

1. Run the workflow in test mode.
2. Send a test voice question through the ElevenLabs agent.
3. Watch the workflow to verify the question is received and answered.

4. Activating for Production

1. Set the workflow to active (triggered mode) so it runs automatically.
2. Consider hosting n8n yourself for better control. Visit self-host n8n for options.
3. Embed the ElevenLabs voice widget on your website for customers to start using it.

Customization Ideas

• Change the Qdrant collection name to use different document sets.
• Modify token chunk sizes for document splitting to affect answer detail.
• Adjust the system prompt in ElevenLabs agent to change the voice assistant’s personality.
• Expand memory window size to handle longer conversations better.
• Connect other document sources like additional Google Drive folders for more knowledge.

Troubleshooting Common Issues

• Webhook does not trigger: Check if ElevenLabs webhook URL matches Webhook node URL. Make sure POST method and body parameters are enabled.
• Qdrant collection not found: Create collection first and double-check spelling of the collection name.
• Embedding or insertion fails: Verify OpenAI API Key is valid. Check document content passed is text format.
• Voice response delayed or missing: Confirm Respond to ElevenLabs node is connected and ElevenLabs agent settings accept responses.
• Google Drive downloads fail: Verify folder or file IDs. Check Google Drive permissions.

Pre-Production Checklist

• Qdrant collection exists and is empty before indexing.
• Google Drive folder has documents and OAuth credentials are correct.
• ElevenLabs webhook triggers properly with test voices.
• OpenAI API Keys are active and have enough quota.
• Documents are vectorized and indexed into Qdrant.
• Test voice questions result in accurate voice answers.

Deployment Summary

Switch the workflow to active in n8n to receive real-time questions.
Embed the ElevenLabs widget on the website for customers to ask by voice.
Monitor workflow logs to fix any errors early.

For more control, consider self-host n8n on your own server.

Summary

✓ The workflow answers voice questions using AI and document search.
✓ It reduces human effort by automating replies in natural speech.
✓ Users get instant, accurate customer answers without delays.
✓ Businesses save time and improve customer service quality.