AI automation · Internal knowledge

KnowledgeAssist — internal knowledge search with citations

An assistant that answers from your own handbook in plain language — and cites the section it came from.

Typical build 4–8 weeksPeople in the loop for anything uncertainYour tools stay where they are

The workflow

  1. Your documentsHandbooks, SOPs, policies, past tickets
  2. Permission checkOnly what this person may see
  3. Secure retrievalThe passages that actually apply
  4. AI reasoningAn answer drawn from those passages only
  5. AnswerPlain language, in the question’s language
  6. Sources + confidenceCited section, or “not covered — referred on”

01 The problem

You know this one.

The pattern, what it costs you, and what changes once it is built.

What happens today
The same five policy questions reach the same person every week, and the answer has been written down for years in a document nobody can find.
Why it matters
Answering is quicker than explaining where to look, so the loop never breaks — and the answer depends on who you ask.
What changes
Questions are answered from the document itself, with the section cited. The ones your documents do not cover finally become visible.
What improves
Senior time back, consistent answers, faster onboarding — and a ranked list of gaps in your documentation.

02 Watch one run

Follow one question from your documents to an answer.

Scroll and the pipeline moves with you. On a phone it plays by itself — tap it, or a stage name, to take control.

Illustrative interface · example data
  1. Ingest

    Your handbook, your SOP folder, the internal wiki, three thousand resolved tickets, the price list somebody updates every week. Each keeps its version date, so an answer can always be traced to the document that was current at the time.

  2. Index

    The step most demos skip. Every passage inherits the read permissions it already had, so payroll guidance stays with HR and supplier terms stay with procurement. An assistant that answers everything to everyone is not a feature, it is a data breach.

  3. Retrieve

    When a question arrives, the relevant passages are pulled and scored. Anything below the threshold is left out rather than padded in — a weak match is how a confident-sounding wrong answer gets built.

  4. Answer

    A plain answer with the section cited, in whichever language the question was asked. And when your documents genuinely do not cover it, it says so and passes the question to whoever owns that policy. Refusing to guess is the whole point.

  5. Improve

    Every question with no source document is logged. Over a month that becomes a ranked list of the gaps in your own documentation, assigned to owners — which is often worth more than the answering itself.

03 The engagement

What you get, and what it takes.

A fixed price is agreed in a written scope before anything is built. The discovery call that comes first is free.

What you get

  • Knowledge-source audit
  • Assistant trained only on your own documents
  • Permission-aware answers and multi-language support
  • Confidence flagging plus a monthly content-gap report

What we need from you

  • The documents you want it to answer from
  • How access is organised today — who may read what
  • An owner for each policy area

Timeline

4–8 weeks

From signed scope to running in your systems, with working software shown to you at every milestone.

How an engagement runs

Connects to tools like these — and anything with an API or export

  • Google Drive
  • SharePoint
  • OneDrive
  • PDF and Word files
  • Internal wikis
  • Helpdesk history

04 Questions

Before you open up your knowledge.

Could it show information to people who should not see it?

Every passage inherits the read permissions it already had, so a question only ever returns answers the person asking is already allowed to see.

What if our documents do not cover the question?

It says so and refers the question to whoever owns that policy. It does not invent an answer.

Where does our data go?

Your documents are used to answer your own team’s questions, and data-handling terms are agreed with you in writing before anything is connected. Our Security page explains what we do today.

05 Often combined

Most engagements use two or three.

The five automations share one engineering spine, so they combine cleanly — and feed the same data into SolmiriX if you use it.

  • DocTrace

    Invoices, forms and delivery notes get read, checked and posted into the systems that own them — instead of retyped by hand. DocTrace

  • SupportRoute

    Email, chat, WhatsApp and voicemail sorted the moment they land — so the urgent one reaches a person while it still matters. SupportRoute

  • ProspectRoute

    Every enquiry captured, scored and assigned to the right rep within minutes — with a clock on it, so nothing goes cold. ProspectRoute

  • UnifyCore

    Two systems that cannot talk, connected in both directions — with retries, a review queue and an alert instead of silence. UnifyCore

Compare all five

Want to see KnowledgeAssist on your own process?

Tell us how the work gets done today. We will tell you plainly whether this is the right fit — and what it would take.