Build a multi-tenant RAG search engine with Qdrant and OpenAI GPT-4o

Description

Quick overview
This workflow exposes a POST webhook that authenticates a tenant, embeds the user query with OpenAI, searches tenant-scoped vectors in Qdrant, and uses GPT-4o to return a grounded answer with bracketed citations and usage metrics.

How it works
Receives a POST request via a webhook with a user query and tenant identifiers (for example, x-tenant-id in headers).
Validates that the tenant ID and query are present and routes invalid requests to an HTTP error response.
Sends the query to OpenAI Embeddings (text-embedding-3-small) to generate a vector.
Builds a Qdrant search request that enforces multi-tenant isolation using payload filters (required tenant_id, optional workspace_id and category) and applies a similarity threshold.
Queries Qdrant for the top matching chunks, then formats the retrieved payload into a single context block and a structured citations list.
Sends the query and formatted context to OpenAI Chat Completions (GPT-4o) to generate an answer that cites sources like [Source 1] using only the provided context.
Assembles a final JSON response (answer, citations, similarity metrics, and token usage) and returns it to the webhook caller.

Setup
Provide an OpenAI API key (used for embeddings and GPT-4o), either by setting OPENAI_API_KEY as an environment variable or replacing the placeholder in the HTTP Authorization headers.
Set up a Qdrant instance and configure QDRANT_HOST, QDRANT_API_KEY (if required), and QDRANT_COLLECTION (defaults to enterprise_rag_vectors).
Ensure your Qdrant collection stores payload fields used for filtering and citations (at minimum tenant_id, and optionally workspace_id, category, plus fields like content/text, title/filename, and document_id).
Copy the webhook URL from the Webhook node and configure your client to POST a JSON body containing query and to include x-tenant-id (and optionally x-workspace-id) headers.

Requirements
• OpenAI API Account (with text-embedding-3-small and GPT-4o access)
• Qdrant Cluster (Self-hosted Docker on VPS or Qdrant Cloud Free Tier)
• n8n instance (Cloud or Self-hosted v1.0+)

Customization
• Swap text-embedding-3-small with local Ollama / BGE embeddings for completely air-gapped on-premise deployments.
• Adjust the cosine similarity threshold (default 0.68) inside the Code node to fine-tune precision vs recall.
• Connect the output response to Slack, WhatsApp, or customer support chatbots for automated enterprise document search.