Skip to main content

RAG knowledge assistant for your company documents

Employees spend a lot of time looking for the current version of a policy, a contract clause or a technical specification. A retrieval-augmented assistant answers these questions from your own documents, shows its sources and respects who is allowed to see what.

ISO 9001 & ISO/IEC 27001Certified quality and information securitySince 2014, made in GermanyDiese Seite auf Deutsch

How a RAG assistant differs from a plain chatbot

Retrieval-augmented generation splits your documents into passages, indexes them and retrieves the most relevant ones for each question before a language model writes the answer. The model is therefore working from your manuals, contracts and wikis rather than from general training data. Quality depends far more on document preparation, retrieval and permissions than on which model you choose.

We build these assistants on kalAIDoCreate, the jocapps platform for enterprises and the public sector with governance, audit logs and custom APIs. Where documents contain personal data that the assistant does not need, Textwash Pro anonymises them locally before indexing, so names, addresses and identifiers never reach the AI system in the first place.

What to watch out for

  • Scanned PDFs, complex tables and slide decks lose meaning if text extraction is poor. Check a sample of each format before indexing everything.
  • If several versions of the same document exist, the assistant may quote an outdated one. Define which repository is authoritative and exclude drafts and archives.
  • Permission filtering must happen at retrieval time, not only in the user interface. Otherwise restricted content can still leak into generated answers.
  • Measure answer quality with a fixed set of test questions and expected sources. Without that baseline, every change to chunking or prompts is guesswork.
  • Employees may use the assistant for personnel or performance questions. Clarify scope, logging and evaluation of usage with the works council where one exists.

What your teams get

Answers with sources

Every answer links to the passages it is based on, including document title and section. Users can verify in seconds instead of trusting a summary blindly.

Permissions respected

Retrieval is filtered by the user's existing access rights, for example from SharePoint, a DMS or group memberships. Someone in sales does not get answers from HR files.

Privacy by design

Textwash Pro removes personal data from documents locally before they are indexed. That reduces the data protection impact and makes approval by your data protection officer easier.

Kept current automatically

Connectors re-index changed and deleted documents on a schedule. Outdated versions drop out of the index instead of producing contradictory answers.

How we build your knowledge assistant

  1. Source inventory

    We map document sources, formats, owners and access rules, and decide which collections go in first.

  2. Preparation

    Documents are cleaned, anonymised where needed and split into passages that keep tables and headings intact.

  3. Retrieval tuning

    Using a set of real questions from your staff, we tune search, ranking and prompts until answers are reliable.

  4. Rollout

    A pilot group uses the assistant first, feedback is reviewed, and access is then widened department by department.

Matching products

kalAIDoCreate

Governed AI platform for enterprises and the public sector. An all-in-one AI platform for enterprises and public institutions with compliance features, governance and custom APIs.

Explore kalAIDoCreate

Textwash Pro

Local, automated text anonymization. Anonymize documents automatically with local machine learning before they reach any AI system. Privacy by design for regulated teams.

Discover Textwash Pro

Frequently asked questions

Which document sources can be connected?

Common sources are SharePoint, file shares, Confluence, document management systems and PDF collections. Other systems can be connected via kalAIDoCreate's custom APIs.

Do our documents leave the EU?

That depends on the model and hosting you choose. kalAIDoCreate supports setups that keep data processing within the EU, and Textwash Pro reduces what personal data is processed at all.

How is this different from enterprise search?

Search returns a list of documents, while a RAG assistant reads the relevant passages and formulates a direct answer with references. Both rely on good indexing, and a RAG assistant can build on existing search infrastructure.

Can the assistant also answer customer questions?

Yes, with a separate, public document collection and stricter limits. Internal and external assistants should never share an index.

Let's talk about your project.

15 minutes with an AI Code Engineer, free and without obligation. Or call our team around the clock.