Give your team an AI knowledge base that answers with its source.
The answer is somewhere in your shared drives, and the one person who knows where is in a meeting. We build a private assistant on your procedures, contracts and past tickets that answers in plain words, with the exact passage it drew from. Each person only gets answers from files they are already allowed to open.
Today
- Shared drives, wikis and old emails to dig through
- The same question asked to the same expert, every week
- An answer found, but from last year’s version
- New hires waiting for someone to be free
With your assistant
- One question, in plain words, from the tools you already use
- An answer drawn only from your documents, source cited
- Each person sees only what they are allowed to see
- When the documents are silent, it says so
What is an AI knowledge base?
An AI knowledge base is an assistant that answers questions from your own documents. It first finds the passages that matter in your files, then a language model writes a short answer from them and cites each one. The technique is called RAG, for retrieval-augmented generation. When the passages do not cover the question, the assistant says it does not know.
It works on the documents your company already has: procedures, contracts, technical manuals, product sheets, HR policies, answered tickets, past proposals. They stay where they are. The assistant reads them, picks up every new version, and applies the access rights already set in your tools.
Type a question into your intranet search and you get a list of files to open. Ask the assistant and you get a short answer and the page it comes from. A public chatbot has never seen your files, so it has nothing to say about them. Since each sentence links to its source, anyone on the team can check it in one click.
Search bar, public chatbot or your own assistant?
Your team probably uses the first two already. Here is what changes with an assistant built on your own files.
| Criterion | Intranet search | Public AI chatbot | Your knowledge assistant |
|---|---|---|---|
| Understands a question in plain words | No, needs the right keywords | Yes | Yes |
| Answers from your documents | Returns a list of files | No, it has never seen them | Yes, and only from them |
| Shows where the answer comes from | The file, you read the rest | Rarely, and never your files | The exact passage, every time |
| Respects who may see what | Yes, if the intranet does | No | Yes, your existing access rights |
| Keeps up when a document changes | Yes | No | Yes, the new version replaces the old one |
| Where your documents go | They stay in-house | To the provider, under its terms | Europe, private cloud or your servers |
The questions your team will stop asking twice.
Knowledge that sits in documents and in a few people’s heads. Any of these areas makes a sensible first assistant.
Customer service
BeforeAgents put the customer on hold to look for the right procedure, or ask a colleague.
AfterThey ask the assistant during the call and read the answer, with the procedure it comes from.
Sales and tenders
BeforeEach proposal starts from a blank page, although the answer was written for a past bid.
AfterThe assistant finds what was already answered, and where, so the team reuses its best work.
Onboarding
BeforeNew hires collect answers from whoever has time, and hesitate to ask twice.
AfterThey ask as often as they need, and learn where the reference documents live.
Field and technical teams
BeforeTechnicians call the office to find a setting buried in a manual.
AfterThey ask from their phone and get the page of the manual, with the value.
Contracts
BeforeFinding a notice period or a penalty clause means reopening every contract.
AfterThe assistant answers with the clause, the contract and the page, for the people allowed to read it.
HR and internal rules
BeforeHR answers the same questions on leave, expenses and remote work all year.
AfterEmployees get the answer from your current policy, and HR keeps the tricky cases.
Built so every answer can be checked.
Security and dataIt only answers from your files.
When nothing in them covers the question, it says so instead of guessing.
Every answer shows its source.
Document, page and passage sit one click away from the answer.
Your access rights apply.
A person only gets answers from documents they could already open in your tools.
Your documents stay where you decide.
Hosted in the EU, in a private cloud or on your servers, with providers and settings that exclude training on your data.
Tested on your real questions.
Before launch, your experts grade its answers to questions your team actually asks.
From your shared drives to a first team, week by week.
Week 1
Choose the documents
We pick the sources where answers are hardest to find and collect the questions your team really asks. The week ends with a one-page scope and a fixed price.
Weeks 2 to 3
Connect and index
Your sources are connected with read-only access, your access rights are carried over, and outdated versions are set aside.
Weeks 4 to 5
Test with your experts
The assistant answers your set of real questions. Your experts grade each answer, and we fix what they flag.
Week 6
Open to a first team
A pilot team uses it every day and rates the answers. Then we open it wider, with a dashboard of what people ask.
The price of a knowledge assistant depends on your documents.
The model is rarely what drives the cost. Where your files live, how clean they are and how finely access is split weigh far more, along with the points below. Once we have looked at a sample of your files, we put the scope and the price in writing, before any work starts.
See market price ranges in the AI agent cost guide- 01How many sources to connect: drives, wiki, CRM, ticketing, email.
- 02The volume and state of the documents: scans, tables, duplicate versions.
- 03How fine-grained your access rights are.
- 04Where your team uses it: chat tool, intranet, inside an existing application.
- 05Where it runs: cloud in the EU, private cloud or your own servers.
- 06How much testing and traceability your sector requires.
Knowledge assistant: the questions buyers ask us
Which documents can the assistant read?
PDF, Word, slides, spreadsheets, wiki pages, emails and tickets, wherever they are stored. Scanned documents go through text recognition first. We check a sample of your real files at the start, because the state of the documents matters more than their format.
Can it make things up?
It is built to avoid it. It only answers from the passages it found in your documents, cites them, and says when your files do not cover the question. No system is perfect, so every answer carries a one-click link to its source, and we test the assistant on your own questions before anyone relies on it.
Will people see documents they should not?
No. The assistant carries over the access rights of your existing tools. A person only gets answers built from documents they could already open, and a document removed from their access disappears from their answers too.
Where are our documents hosted?
Where you decide: in a European cloud, in your private cloud or on your own servers. We use providers and settings that exclude training on your data, and models from Mistral or open-source models when nothing may leave your walls.
What happens when a document changes?
Your sources are synchronized on a schedule set with you. The new version replaces the old one in the answers, and documents marked as archived stop being used, so nobody is sent to last year’s procedure.
How long does it take to go live?
A first assistant on a well-bounded set of documents usually goes live in four to eight weeks, depending on the number of sources and their state. The first week always ends with a written scope and date.
Knowledge assistant, by specialty
Often built together
- Try the knowledge assistant demo Ask questions of sample documents and see the sources.
- AI chatbot development When your customers need answers too.
- What is RAG? The technique behind it, in plain words.
- AI agents When the answer should turn into an action.
- Document AI When the data is stuck in scans and PDFs.
- Private LLM When nothing may leave your servers.
Which question does your team ask every single week?
Name the documents your team keeps digging through. Within one business day, we tell you which questions an assistant could answer from them and how we would start. The 30-minute assessment is free.