Multi-Tenant RAG Search n8n Workflow: Qdrant + GPT-4o

Multi-Tenant RAG answers from Qdrant + GPT-4o—via a single n8n POST webhook

Bring Retrieval-Augmented Generation (RAG) to your SaaS: this n8n workflow receives a tenant-scoped query, embeds it with OpenAI, searches the correct Qdrant vectors, and returns a grounded GPT-4o answer with bracketed citations and usage metrics—without leaking data across tenants.

What this workflow does

• Exposes a POST webhook that accepts a user query plus tenant identifiers (e.g., x-tenant-id in headers).
• Validates inputs (tenant ID and query). Missing/invalid requests are routed to an HTTP error response.
• Creates a vector embedding using OpenAI Embeddings (text-embedding-3-small).
• Performs multi-tenant Qdrant search with payload filters enforcing tenant_id, plus optional workspace_id and category, along with a similarity threshold.
• Formats retrieved chunks into a single context block and builds a structured citations list from the payload.
• Generates a grounded response using OpenAI Chat Completions with GPT-4o, instructing it to use only the provided context and cite sources like [Source 1].
• Returns a final JSON response including answer, citations, similarity metrics, and token usage.

Use cases

• Tenant-specific support Q&A where each workspace’s knowledge base must remain isolated.
• SaaS operators building an internal “ask your docs” feature with Qdrant-backed RAG.
• Automation engineers deploying a webhook-driven RAG endpoint for chat or agent workflows.

Technical details

• n8n nodes: webhook, if, code, http request, respond to webhook, plus a sticky note for guidance.
• Integrations: OpenAI (embeddings + GPT-4o) and Qdrant (vector search with multi-tenant payload filters).
• Environment/config: set OPENAI_API_KEY, and configure Qdrant connection details such as QDRANT_HOST and QDRANT_API_KEY if required.