n8n ships with a Qdrant Vector Store node, which means you can lay out the entire RAG pipeline — document to embedding to Qdrant to an AI answering from your data — without writing a line of code. If your n8n also runs on RoamerHost, both services sit in the same data center and internal latency is extremely low.
Before you start
- An n8n instance (RoamerHost or self-hosted, either works)
- A Qdrant instance: note the endpoint and API key once it's provisioned
- An API key for an embedding model (OpenAI or another provider)
1. Create the Qdrant credential in n8n
In n8n, go to Credentials in the left menu and add a "Qdrant API" credential: put your endpoint in Qdrant URL (https://qdr-xxxx.roamerhost.com) and the key shown in the dashboard in API Key. n8n tests the connection automatically when you save.
2. Build the document ingestion workflow
- Trigger: use a Form Trigger (upload files manually) or a Google Drive Trigger (process new files in a folder automatically)
- Qdrant Vector Store node: set the mode to "Insert Documents" and name the collection whatever you like (for example
company-docs) - Attach an Embeddings sub-node (OpenAI Embeddings, for instance) and a Text Splitter sub-node (a chunk size of 500–1000 with 50–100 overlap is a good starting point)
Run the workflow once and your documents are split, embedded and written into Qdrant.
3. Build the AI question-answering workflow
- Trigger: a Chat Trigger (n8n's built-in chat interface) or a webhook (to connect your own website, LINE or Telegram)
- AI Agent node: pick whichever LLM you normally use
- Add Qdrant Vector Store to the agent's Tools (mode set to "Retrieve Documents (As Tool)", pointing at the same collection)
Once it's done, ask it something about your company and the agent will fetch the relevant passages from Qdrant before answering, so the answer is grounded. Ask about something that isn't in the data and it will say it doesn't know rather than making something up.
Common problems
Search results aren't accurate? First check whether ingestion and querying use the same embedding model — different models mean different coordinate spaces, and the search cannot be accurate. Then adjust the chunk size: chunks that are too large add noise, chunks that are too small lose the meaning.
How do I update documents? Just run the ingestion workflow again. To avoid duplicates, clear the old data first using the Qdrant node's delete mode, or record a document version in the payload.
Want someone to build it for you?
If you'd rather not assemble these workflows yourself, or the scale is big enough that you want someone planning it with you, Roamer Tech (RoamerHost's parent company) takes on contract work in business process automation and AI agents:
