PromptPartner DE

Knowledge management

Who does your team ask when the answer is somewhere in the files?

An AI agent answers from your manuals, SOPs, contracts and minutes, with the source next to every answer. Your people check what it says, and anything that leaves the house is approved by one of them first.

Four colleagues at a table with a laptop, a process drawn on the whiteboard behind them
Answer: torque, flange X200Illustration with sample data
  1. 45 Nm, M10 bolts, tightened crosswiseManual X200, page 84
  2. Value changed in 2024Change note 24-117
  3. Same fix on a machine in serviceService report 2025-311
  4. Open: spare part numbernot found in the documents

Checked by your service team

Most answers are written down somewhere, in places only a few people know.

  • The manual, the SOP and last year's quote sit in three folders with three naming rules.
  • New colleagues ask the same two people, and those two answer the same questions every week.
  • When the person who knows is on holiday, the question waits.
  • The minutes record what the board decided, but finding out why means reading through years of them.

Three things we build from your documents.

Each one reads only the documents you choose, and every answer shows where it comes from.

  • An assistant that answers with sourcesYou ask in your own words and get an answer with a link to the page it comes from. The method is called RAG: the agent first looks up the passages that match, then writes the answer from them.

  • A knowledge graph for questions across documentsA model reads all your documents beforehand and notes what they mention and how it connects: supplier, machine type, deviation. This is called GraphRAG, and it answers questions such as which fault keeps coming back.

  • One search across your sourcesSharePoint, Confluence, Google Drive and other sources in one search, as far as the chosen tool connects to them.

If your software can already do the job, or your documents can't carry it, we tell you and stop there.

From your documents to an answer one of your people can check.

The agent reads the documents you name and builds a search index from them and, where it pays off, a knowledge graph. It then answers with the source next to each statement.

  • Manuals
  • SOPs
  • Service reports
  • Contracts and minutes

The agent

Answers from your documents

search indexknowledge graph

Answer with sourcesExample

One of your people checksNothing goes out without a yes

Underneath, always

  • AccessOnly the documents each team may see.
  • GuardrailsRules it works within.
  • LogEvery step is written down.
The agent answers from your documents and shows its sources. Illustration with sample data.

When is a knowledge graph worth it?

Most everyday questions have their answer on one or two pages, and search with sources is enough for those. A graph costs more to build and has to be refreshed when your documents change, so we add it only where your questions need it.

Search with sources (RAG)

For questions whose answer sits on one or two pages.

  • What torque does the manual give for model X200?
  • What does our expense policy say about train travel?
  • Which SOP applies to this step?

In Microsoft 365 this is Copilot or an agent in Copilot Studio. In your dedicated environment we use AnythingLLM, which shows the source next to each answer.

Knowledge graph (GraphRAG)

For questions across many documents and the links between them.

  • Which supplier is linked to the deviations on line 3?
  • Which fault keeps coming back on this machine type?
  • Which themes recur in five years of board minutes?

In your dedicated environment we run LightRAG for this, with Neo4j as the graph database where needed. For a custom build we use Microsoft's open-source GraphRAG library.

Custom agents and apps

Sources: Edge et al., From Local to Global: A Graph RAG Approach (arXiv, 2024); Microsoft GraphRAG on GitHub; Alexander Thamm, From RAG to GraphRAG.

Every trade keeps its knowledge in different documents.

  • Machine and plant buildersManuals, service reports and spare-part lists, for the service team on the phone.

  • Life sciencesSOPs and deviations, with the SOP section next to each answer.

  • ConstructionStandards, tender documents and contracts, for the question of what the tender requires.

  • Technical wholesaleProduct data, supplier catalogues and certificates, for the sales back office.

  • Finance back officePolicies, procedures and contracts, such as the notice period in a contract.

  • Board and managementPast minutes and decisions: when something was decided, and why.

In your Microsoft 365, or in your dedicated environment, managed by us.

In your Microsoft 365

With the Microsoft 365 Copilot (Premium) licence, Copilot answers from your mail, files and meetings, and each person only sees what their own rights allow.

For a knowledge agent of its own, we build it in Copilot Studio with your SharePoint sites as its knowledge. It only shows each person content that person can access, and we set it up so that an answer from your documents comes with a citation.

Because Copilot shows each person everything their rights allow, we first check with your IT who can open which SharePoint sites. The Premium licence includes SharePoint Advanced Management, which helps reduce oversharing.

In Microsoft 365

In your dedicated environment, managed by us

We run open-source knowledge tools for you on AI Admin Panel, the platform we build: AnythingLLM for answers with sources, Open WebUI as a team chat with its own knowledge bases, and LightRAG for the knowledge graph.

For one search across SharePoint, Confluence and Google Drive, we set up Onyx in its Standard version, which connects to them. For its answers to follow each person's rights in those sources, Onyx needs its Enterprise Edition.

Secret files, such as drawings or recipes, go only to a private model in that environment. One control panel shows all your agents, apps and AI models, including your private ones.

Your people sign in with your Microsoft Entra ID, Google or Okta account, and others on request. Through a tailnet, with Tailscale, Headscale or NetBird, the agents and models we host for you connect to your network as if they were in your own.

Managed AI environmentPrivate AIOpen-source agents

Who checks the answer?

The agent prepares the answer. What happens with it is decided by someone on your team.

  • Every answer shows its sourcesEach statement links to the page it comes from, so the person asking can check it before relying on it.

  • It says when it finds nothingWe set the agent up to say so when your documents hold no answer to the question.

  • A person decides what goes outAnswers for your own team are there to read. Anything that goes to a customer, an authority or the board is checked and approved by one of your people first.

  • Access follows your rulesIn Microsoft 365 the agent follows the rights you have already set. In your dedicated environment, each team gets a workspace with only the documents it may see.

Which documents should we start with?

Knowledge management has no trial run. You start with the first call, and larger firms with the AI assessment.

  1. 01

    First call

    We look at where your documents sit and whether one set of them suits an agent.

  2. 02

    AI assessment

    At the end you have a fixed-price proposal or a clear no with reasons.

  3. 03

    The agent at work

    The monthly fee starts only once the agent passes acceptance on your past cases.

Questions about knowledge management

Can we trust the answers?

Every answer shows the page it comes from, so the person asking can check it before relying on it. We set the agent up to say so when your documents hold no answer. Anything that goes to a customer, an authority or the board is checked and approved by one of your people first.

Will people see files they are not allowed to open?

In Microsoft 365, Copilot and Copilot Studio agents show each person only content that person can access. In your dedicated environment, each team gets a workspace with only the documents it may see. If answers must follow each person's rights in SharePoint, Confluence or Google Drive, we tell you before you decide which tool fits.

Do we need a knowledge graph?

Usually not at first. Most everyday questions have their answer on one or two pages, and search with sources is enough for those. A graph costs more to build and has to be refreshed when your documents change, so we add it only where your questions run across many documents.

We already have Microsoft 365 Copilot. Why would we need this?

With the Premium licence, Copilot already answers from your mail, files and meetings. We add a knowledge agent for a set of documents, check with your IT who can open which SharePoint sites, and keep a second way open in your dedicated environment. If Copilot already does enough, we tell you.

Where do our documents go?

They stay in your Microsoft 365, or they go to your dedicated environment, managed by us, which a tailnet connects to your network. Secret files, such as drawings or recipes, go only to a private model in that environment.

Which question does your team ask most often?

In 30 minutes we look at where your documents sit today and which of the two places suits them.