n8n RAG Document Chatbot: Google Drive + Gemini + Qdrant

Auto-chunk, embed, and store Google Drive documents in Qdrant, then converse via a Gemini-powered AI Agent with retrieval augmentation

Who

Operators or developers who need to build a Q&A bot over private document repositories

You get

Documents from a Google Drive folder are processed and stored in a vector database; users get answers through a chat interface grounded in retrieved context

You need

n8n · google-drive · qdrant · gemini · telegram · google-docs

Steps

1. Configure API Credentials

   Create and link the following credentials in n8n: Google Drive OAuth2, Google Docs OAuth2, Google Gemini (googlePalmApi), OpenAI API (for text-embedding-3-large and the delete script), Qdrant API, and Telegram Bot. Enter the OpenAI API Key in the 'Delete Qdrant Points by File ID' code node configuration.

2. Configure Document Sources and Collection

   Enter the target Google Drive folder ID in the 'Google Folder ID' node. Define the collection name in the 'Qdrant Collection Name' node (currently hardcoded as 'nostr-damus-user-profiles' in the template; update to match your document scenario).

3. Batch Retrieve and Download Documents

   Run the 'Find File Ids in Google Drive Folder' node to list files, aggregate via 'File Id List', then process in batches through 'Loop Over Items'. Each batch is downloaded via 'Download File From Google Drive' as binary data.

4. Extract Metadata and Split

   'Get File Contents' converts files to text, then 'Extract Meta Data' uses Gemini to extract structured attributes (theme, pain points, keywords, etc.). 'Token Splitter' chunks at chunkSize=3000, and 'Data Loader' attaches extracted metadata to document metadata.

   Execution prompt · Source text。Extract Meta Data node's systemPromptTemplate field

   Copy prompt

   You are an expert extraction algorithm.
   Only extract relevant information from the text.
   If you do not know the value of an attribute asked to extract, you may omit the attribute's value.

5. Embed and Write to Qdrant

   'text-embeddings-3-large' generates vectors, which are written with chunked documents via 'Qdrant Vector Store' in insert mode. A 'Wait' node paces batch processing, and 'Send Completed Message' notifies via Telegram.

6. Start the Chat Trigger

   Enable the 'When chat message received' Chat Trigger node as the conversation entry point. It passes user input to the AI Agent.

7. RAG Conversation and History Logging

   The 'AI Agent' calls 'Qdrant Vector Store Tool' in retrieve-as-tool mode (topK=20) to fetch relevant snippets, then generates answers via 'Google Gemini Chat Model1'. 'Window Buffer Memory' maintains a 40-turn context window. Final output returns through 'Respond to User', while 'Update Chat History' appends Q&A to a Google Doc.

   Execution prompt · Source text。AI Agent node's systemMessage field (promptType=define)

   Copy prompt

   =You are an intelligent assistant specialized in answering user questions using Nostr user profiles. Your primary goal is to provide precise, contextually relevant, and concise answers based on the tools and resources available.
   
   ### TOOL
   Use the "nostr_damus_user_profiles" tool to:
   - perform semantic similarity searches and retrieve information from Nostr user profiles relevant to the user's query.
   - access detailed information about Nostr and/or Damus users when additional context or specifics are required.
   
   ### Key Instructions
   1. **Response Guidelines**:
      - Clearly explain how the retrieved information addresses the user's query, if applicable.
      - If no relevant information is found, respond with: "I cannot find the answer in the available resources."
   
   2. **Focus and Relevance**:
      - Ensure all responses are directly aligned with the user's question.
      - Avoid including extraneous details or relying solely on internal knowledge.

8. Secure Vector Deletion (Optional)

   To clean up data, run the deletion branch: 'Confirm Qdrant Delete Points' sends a Telegram confirmation request; upon approval, 'Delete Qdrant Points by File ID' executes code to delete vector points matching the file_id. Decline triggers 'Send Declined Message'.

   Execution prompt · Source text。Delete Qdrant Points by File ID code node's execute.code field

   Copy prompt

   const { QdrantVectorStore } = require("@langchain/qdrant");
   const { OpenAIEmbeddings } = require("@langchain/openai");
   
   // Qdrant connection details
   const url = "http://localhost:6333/";
   const apiKey = "";
   
   // OpenAI API configuration
   const openAIApiKey = "[Your-OpenAI-API-Key]";
   
   // Get input data
   const items = this.getInputData()[0];
   // console.log(items)
   
   const collectionName =  items.json.qdrant_collection_name;
   // console.log(collectionName)
   
   async function deleteDocumentsFromQdrant() {
     try {
       // Initialize OpenAI embeddings
       const embeddings = new OpenAIEmbeddings({
         model: "text-embedding-3-large",
         openAIApiKey: openAIApiKey
       });
   
       // Connect to existing Qdrant collection
       const vectorStore = await QdrantVectorStore.fromExistingCollection(embeddings, {
         url: url,
         apiKey: apiKey,
         collectionName: collectionName,
       });
   
       const fileIds = items.json.appended_id.map(item => item);
   
       console.log(fileIds)
   
       // Delete points by fileId
       const deletionResults = await Promise.all(fileIds.map(async (file_id) => {
         const filter = {
           must: [
             {
               key: "metadata.file_id",
               match: { value: file_id }
             }
           ]
         };
   
         try {
           // Access the underlying Qdrant client to perform the deletion
           await vectorStore.client.delete(collectionName, { filter });
           return { file_id, status: "deleted" };
         } catch (error) {
           console.error(`Error deleting documents for fileId ${file_id}:`, error);
           return { file_id, status: "error", message: error.message };
         }
       }));
   
       return deletionResults;
     } catch (error) {
       console.error("An error occurred during the deletion process:", error);
       return error.message;
     }
   }
   
   // Execute the deletion process
   try {
     const result = await deleteDocumentsFromQdrant();
     console.log("Deletion process completed:", result);
     return [];
   } catch (error) {
     console.error("Failed to execute deletion process:", error);
     return [{ json: { error } }];
   }

You are setting up an n8n RAG document chatbot. First confirm these inputs and permissions: 1) Google Drive folder ID and OAuth2 credentials; 2) Qdrant instance URL and API Key; 3) Gemini (googlePalmApi) and OpenAI embedding API Key; 4) Telegram Bot Token and Chat ID; 5) Google Docs document ID for chat history. If importing this template, check whether the AI Agent's systemMessage and Qdrant collection name still contain stale Nostr hardcoding and update them for your document scenario.