Give your team fast, accurate answers from your knowledge base. The flow understands each question, chooses the best strategy, finds the right documents, and writes a clear reply. Ideal for support desks, internal FAQs, and onboarding help.

A chat event or another workflow starts the run. A Google Gemini agent labels the question as factual, analytical, opinion, or contextual. A Switch node then sends it to a matching strategy that tightens the query, breaks it into sub questions, maps perspectives, or adds user context. The system searches a Qdrant collection using Gemini embeddings. It builds a clean context block, then a final Gemini agent uses that context, the user question, and shared chat history to draft the answer, which returns by webhook.

Set your Qdrant vector store id, add Gemini and Qdrant credentials, and define a chat memory key. Teams usually cut response time from minutes to seconds while keeping answers consistent. Use it for customer portals, agent assist, internal IT help, and sales knowledge checks. Expect fewer escalations, better first reply quality, and easier scaling as question volume grows.

• Multi entry triggers with chat interface or calls from another workflow
• Gemini powered query classification into factual, analytical, opinion, or contextual
• Strategy branches that refine the query, create sub questions, list perspectives, or infer user context
• Dedicated chat memory per strategy and a shared memory for the final answer
• Gemini embeddings to search a Qdrant vector store selected by vector store id
• Context concatenation that merges top document chunks into one clean block
• Final Gemini answer uses the system prompt, retrieved context, user query, and chat history
• Webhook response sends the answer back to your app or chat client

• Reduce manual research from 15 minutes to 2 minutes per question
• Automate up to 80 percent of repetitive support answers
• Improve answer precision by about 30 percent with strategy based prompts
• Handle more tickets without extra staff by routing queries to the right strategy
• Connect Google Gemini and Qdrant for unified knowledge delivery

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 Gemini and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. Create Google Gemini credentials: double click any Gemini node such as Gemini Classification, open the Credential to connect with menu, click Create new credential, then follow the on screen steps. If asked for an API key, get it from your Google AI Studio account and paste it in. Give the credential a clear name.
4. Set up Qdrant credentials: double click Retrieve Documents from Vector Store, open the Credential to connect with menu, click Create new credential, then add your Qdrant API key and base URL from your Qdrant dashboard. Name and save the credential.
5. Confirm the Embeddings node uses your Gemini credential. If not, select the same Google Gemini credential you created earlier.
6. Open the Combined Fields node and set defaults for vector_store_id and chat_memory_key. Make sure user_query maps to the incoming question field from your trigger.
7. Check the Switch node rules. The outputs should match Factual, Analytical, Opinion, and Contextual exactly. Edit if your classifier uses different labels.
8. Load your documents into your Qdrant collection. Use the same embedding model family to avoid dimension mismatches. Verify the collection name matches your vector_store_id value.
9. Open Retrieve Documents from Vector Store and set the collection, top results, and any filters. Run a test to confirm documents are returned.
10. Test the chat path: enable the Chat node, open the chat interface in n8n, and ask a question. Watch the execution to see which branch runs and which documents were retrieved.
11. Test the workflow trigger path: use When Executed by Another Workflow to pass user_query, chat_memory_key, and vector_store_id. Confirm the final Answer node returns text in Respond to Webhook.
12. Troubleshoot: if no results, check vector_store_id and Qdrant credentials. If classification feels off, adjust the system message in the Query Classification node. If answers repeat, reduce the memory window length. If responses are slow, lower the number of retrieved chunks.