Turn your company PDFs into fast answers. New files in a Google Drive folder become part of a searchable knowledge base, and a chat bot responds with clear answers. This is great for support and operations teams that need trusted information on demand.

An app event watches a specific Google Drive folder every minute. When a PDF appears, the file is downloaded, the text is extracted, and the content is split into 700 character chunks with a 60 character overlap. Cohere multilingual embeddings convert each chunk into vectors and the records go into a Milvus collection. A chat webhook takes incoming questions, the agent uses OpenAI 4o with conversation memory, pulls the top 10 matching passages from Milvus, and writes a helpful reply based on your documents.

Setup is simple. Connect Google Drive, Milvus, Cohere, and OpenAI accounts, choose your Drive folder and collection name, and keep the default chunk settings unless you need larger or smaller pieces. Expect faster responses, less manual searching, and new PDFs ready for chat in minutes. Common uses include product manuals, policy libraries, onboarding guides, and internal FAQs. Teams cut answer time from minutes to seconds and keep information consistent across the company.

• Google Drive trigger checks a chosen folder every minute for new files
• PDF text extraction converts uploaded files into clean text
• Chunking splits content into 700 character parts with 60 character overlap
• Cohere embed multilingual v3.0 creates vectors for many languages
• Milvus vector store saves and retrieves embeddings by collection
• Top 10 passage retrieval feeds the agent with relevant context
• OpenAI gpt 4o powers the chat responses with a memory buffer
• Chat webhook receives questions and routes them to the RAG agent

• Reduce manual document lookups from 20 minutes to under 1 minute
• Automate 100 percent of PDF ingestion from a Google Drive folder
• Handle up to 10 times more questions with the same support team
• Improve answer relevance using the top 10 matching passages from your files
• Connect Google Drive, Milvus, Cohere, and OpenAI in one flow

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Google Drive, Milvus, Cohere and OpenAI. See the Tools Required section above for links to create accounts with these services.
3. In n8n, double click the Google Drive Trigger node. On the 'Credential to connect with' field, click 'Create new credential' and follow the on screen steps to connect your Google Drive account.
4. In the Google Drive Trigger node, choose the target folder, set the event to fileCreated, and keep the polling schedule at every minute.
5. Open the Google Drive Download node and confirm the File ID is set to {{$json.id}} and the operation is download.
6. Open the Extract from File node and confirm the operation is pdf. Run a test after adding a sample PDF to verify text is extracted.
7. Open the Set Chunks node and confirm chunk size is 700 and overlap is 60. Adjust only if your documents require different sizes.
8. Double click the Cohere embeddings node. Click 'Create new credential', paste your API key from the Cohere dashboard, and select the embed multilingual v3.0 model.
9. Double click the Insert into Milvus node. Click 'Create new credential' and follow the on screen instructions to enter your Milvus endpoint and token. Set the collection name and leave clearCollection off.
10. Open the Retrieve from Milvus node and select the same collection. Keep topK at 10 and confirm the tool name is vector_store.
11. Double click the OpenAI 4o node. Click 'Create new credential', enter your OpenAI API key, and select the gpt 4o model.
12. Open the RAG Agent node and confirm the language model is the OpenAI 4o node, the Memory node is connected, and the vector store tool is enabled.
13. Open the When chat message received node and copy the Webhook URL. Turn on the workflow, upload a PDF to the watched folder, wait one minute, then send a test question to the webhook.
14. Validate results by asking a question covered in the PDF. If answers are weak, check that vectors exist in the Milvus collection, the Cohere key is valid, and the PDF text was extracted.
15. Troubleshoot common issues: ensure n8n has permission to read the Drive folder, confirm the same collection name is set in both Milvus nodes, and verify the file is a readable PDF and not just an image scan.