A RAG chatbot that answers from your own documents

What you use a RAG chatbot for

• Answer employee questions from manuals and procedures
• Draft replies to customer questions from your own knowledge base
• Search contracts, quotes and tickets by meaning
• An internal helpdesk that points to the source
• Add new documents automatically as soon as they are in your folder

How the integration works

The chain consists of fixed steps: loading documents, splitting them into chunks, converting them to vectors and storing them, and for a question, retrieving the best chunks and having the answer written. Qdrant or pgvector run on your own server; the language model can be local or in the cloud. Every answer names the source, and without a fitting source the AI says it does not know.

Facts about the a knowledge base AI API

Checked on 2 November 2026 against the official documentation. Limits, prices and versions change; the source is listed per row.

Fact	What the documentation says	Source
Authentication	Self-hosted Qdrant is open by default, without a key. I always enable an API key, a separate read-only key and TLS. Qdrant Cloud works with database keys per cluster.	Qdrant: Security
Key lifetime	Self-hosted: keys do not expire. Cloud: 90 days by default, configurable.	Qdrant Cloud: Authentication
Rate limits	No API limit in the documentation; the limit is your server.	Qdrant: Interfaces
Interfaces	REST on port 6333 and gRPC on 6334, with official clients for Python, JavaScript, Go and more.	Qdrant: Interfaces
Test environment	The Qdrant Cloud Free tier (1 GB RAM, 4 GB disk) is paused after a week of inactivity and deleted after four weeks. Self-hosted through Docker.	Qdrant Cloud: Create a cluster
Cost	Free self-hosted (open source). Cloud based on usage.	Qdrant Cloud
n8n	Built-in Qdrant Vector Store node: insert, retrieve and use as a tool for an AI agent.	n8n documentation

What you need on your side

• The documents in a readable form (PDF, Word, text) and the folder or system they come from
• A list of who may see which documents
• A server in the EU for the vector database, or a Qdrant Cloud account
• A choice of language model: local or in the cloud

Where it often goes wrong, and how I prevent it

• Outdated documents: an AI is only as good as the documents behind it. Old versions are replaced, not stacked.
• Permissions: not everyone may see everything. The knowledge base respects who may consult which documents.
• Chunking: chunks that are too large or too small give poor answers. I tune that to your type of documents.
• Answer without a source: an answer without a source is a guess. The AI must name the source or say it does not know.

In practice

I already build RAG systems in production. The building blocks are always the same; the difference is in which documents go in and who may see what.

What it costs

I build a knowledge-base AI as one scoped process for a fixed price, from €750. Maintenance and monitoring are a monthly fee: n8n management from €250 per month, server management €145 per server. In the free assessment we work out your process. All prices and what is included →

What I do not do

• I do not put a vector database on the internet without a key and TLS.
• I do not build a knowledge base without a source reference per answer.
• I do not write or correct your documents; outdated content remains your responsibility.

Frequently asked questions about a RAG chatbot

What is a vector database?

A database that stores text as numbers that capture the meaning. That lets the AI find chunks about the same thing, even when they use different words.

Do my documents stay private?

Yes. The vector database runs on your own server in the EU. With a local language model, the documents do not leave your environment.

Can the AI give a wrong answer?

It can, and that is why every answer names the source and the AI says "I don't know" when there is no fitting source. For important answers, a person checks along.

How secure is Qdrant?

As secure as you configure it. Self-hosted Qdrant is open by default, without a key. That is why I always enable an API key, a separate read-only key and TLS, and the database does not listen on the internet.

Should I use Qdrant Cloud or self-host?

Self-hosting on your own EU server suits sensitive documents and costs only the server. Qdrant Cloud has a free entry tier for trying things out, but it is paused after a week of inactivity. For production, self-hosting is my default.