Internal AI assistant with source citations for your own documents
Companies with extensive internal documents whose staff spend a lot of time searching, such as consultancies, fiduciaries, engineering firms or administrations.
Updated: 2026-10
An internal AI assistant answers questions based on your own documents — manuals, policies, contracts, project files — and cites the source for every answer. What matters are permissions, so each person only sees what they may see, evaluation of answer quality and a deliberate choice of model and data residency.
Knowledge sits in documents nobody knows completely: manuals, policies, templates, project files. Staff ask colleagues or search for a long time, and new joiners need months to learn where things are.
General AI tools help little: they do not know your documents, invent answers when unsure and raise questions about confidentiality and data residency.
What you get
Use case and data scope
Defining which documents are included, who may see which content and which questions the assistant should answer.
Retrieval with permissions
Preparing documents with suitable chunking and metadata, and search that filters by permission before content reaches the model.
Answers with sources
Every answer points to the documents used; without a suitable source the assistant does not pretend to answer.
Evaluation
A set of realistic questions with expected answers to measure quality and check it after changes.
Operations and data residency
Choosing model provider and hosting by data protection requirements, cost limits, logging to the permitted extent and a plan for keeping documents current.
How it runs
- 1
Clarification
Discuss documents, permissions, typical questions and data residency requirements.
- 2
Prototype
A limited data scope is prepared and tested with real questions.
- 3
Expansion
Extension to the full scope, permissions, evaluation and integration into existing tools.
- 4
Rollout
Training, AI usage rules and handover with quality measurement in operation.
Price framing
Fixed price by scope. The range depends on document volume, permission model and integrations and is set in a short initial conversation.
Indicative, not a binding quote — the frame is confirmed during scoping.
What drives the price? → Cost guide
Parent service: Custom Software Engineering
FAQ
Are our documents used to train models?
No. We use providers and configurations where training on customer data is contractually excluded, and document the data flows.
Can the assistant give wrong answers?
The risk can be reduced considerably but not eliminated. That is why every answer cites its source, and without a suitable source the assistant does not pretend to answer.
Where is the data?
That is decided with you, for example data residency in Switzerland or the EU, aligned with your requirements and those of your clients.
Does the assistant fall under the AI Act?
An internal knowledge assistant is usually not high-risk. AI literacy and, where applicable, transparency obligations apply, which we take into account at rollout.
Internal AI assistant with source citations for your own documents
Companies with extensive internal documents whose staff spend a lot of time searching, such as consultancies, fiduciaries, engineering firms or administrations.