Answers from your documents — with a cited source
Your policies, contracts, manuals and tickets in one place — with an answer whose source you can always see. (Technically: RAG — retrieval-augmented generation.)
When companies call us
- Knowledge is scattered across SharePoint, Drive, Notion and the heads of two people.
- A new colleague asks what the previous one asked — the answer exists, nobody can find it.
- You uploaded a PDF to a chatbot and it started inventing numbers.
- Legal or security stopped the project because nobody could say who can access what.
The difference between a demo and a usable system is what happens when the answer is not in the documents. A weak system invents it. A good system says “I do not have this in the sources” — and that is a feature, not a defect.
We build RAG that attaches a concrete source to every answer, respects the access permissions from your system, and records who asked what. Without the last part, compliance will not sign it off.
We do not judge quality by impression. We assemble a set of real questions from your company, measure accuracy and show you the numbers — including where the system fails.
How we work
1. Source inventory
We find out where the data actually lives, what state it is in and what is out of date. This phase tends to be uncomfortable and is the most important one.
2. Permission-aware indexing
Documents are processed so the system remembers who may see them. A user never receives an answer from a source they cannot access.
3. A test set of questions
Together with your team we collect 50–150 real questions with correct answers. This is the only way to measure improvement.
4. Deploy where people actually work
Into Slack, Teams, your intranet or your own app. A knowledge base with a separate login never gets adopted.
What you get
- Search and chat over your documents, citing a specific source with every answer.
- Connections to existing storage — SharePoint, Google Drive, Confluence, Notion, databases.
- Document-level access permissions honoured at retrieval time.
- An audit log of questions and answers for compliance.
- Measured accuracy on your test set and a plan for improving it further.
Indicative scope
- Source inventory and architecture design
- €1,200 – €2,400
- Pilot over a single data source
- €4,000 – €8,000
- Production rollout with permissions and audit
- from €9,000
- Operations, reindexing and evaluations
- from €450 / month
Prices are indicative and exclude VAT. The state of your data affects the price far more than the number of documents.
Frequently asked questions
Why is uploading documents to ChatGPT not enough?
For personal use it often is. A company deployment needs three more things: access permissions (not everyone may see everything), citations (you must be able to verify an answer) and audit (you must be able to prove what the system told whom).
What if our documents are outdated or contradict each other?
That surfaces in the first phase and it is the most common finding. The system can flag contradictions and prefer newer sources, but good retrieval does not fix bad content. A cleanup list is part of the deliverable.
How do you measure whether it works?
On a test set of questions from your company. We track how many answers are factually correct, how often the system correctly says “I do not know”, and how often it cites the wrong source.
Can it run without sending data outside the company?
Yes. We can deploy a smaller model on your infrastructure. Answers tend to be somewhat weaker than with the largest models — we measure that gap on your data so you can decide based on numbers.
If your people spend time hunting for information that already exists in the company, this is one of the fastest-paying AI projects there is.