Knowledge Base — semantic search over what your business knows

Semantic search over your documents and records — answers respect each person’s permissions, and your AI assistant cites the same knowledge base.

Part of AxisIQ's single business model — The knowledge layer (Understand the business). How the layers fit.

Every business runs on knowledge that sits in no record: the returns policy, how you grade seconds, why one customer gets longer payment terms, what to say when a supplier misses a date. AxisIQ's Knowledge Base makes that searchable by meaning — you upload the documents, switch on the record types worth including, ask in ordinary words, and every result you get back is filtered by what you are allowed to see.

That last clause is the whole design. A search tool that quietly surfaces the salary sheet because it happened to sit in the same folder is worse than having no search tool at all. Here, permission is decided per result at the moment you ask, not assumed when the file was uploaded.

Semantic search matches your question to the material by meaning rather than by shared words. Ask "what do we do when a parcel turns up damaged" and the returns document comes back even if it never contains the word damaged, because it is about the same thing. That is the question a keyword box cannot answer and this can.

What a company knows is not all the same kind of knowledge, so the base keeps three scopes apart and lets you search across whichever ones you choose.

  • Platform — AxisIQ's own product documentation, available to every member of your workspace. "How do I add a stage to a queue" gets answered from inside the product, not from a browser tab on somebody else's site.
  • Organisation — what your team uploads, filed into named collections, plus any record types you have opted in.
  • Personal — your own notes, tied to your account and visible to nobody else, the workspace owner included.

Keeping them separate is deliberate. Your written operating procedure and your half-formed notes about a supplier are both useful, and they should not sit in the same pile or reach the same audience.

Your records count as knowledge, if you say so

Documents are the obvious half. The less obvious half is that most of what a business knows has already been typed into it — the note on a customer account, the specification on a product, the reason someone wrote on a credit note.

Record indexing is opt-in per object type. Whoever manages the knowledge base switches on the types that should be searchable by meaning, and from that point their records answer questions alongside the uploaded documents. When a record changes it is re-indexed in the background, so what you search is the current state of the system rather than a copy somebody remembered to refresh.

The opt-in matters. It keeps high-volume, low-meaning types out of your results, and it makes exposing a type a decision someone made rather than a default nobody chose. Operations covers how those object types get defined in the first place.

Permission is checked on every single result

Most knowledge tools treat access as a property of the folder. This one treats it as a question asked fresh on every hit, against the person searching:

  • A document in a collection you cannot query never appears in your results.
  • A record you cannot read never appears in your results.
  • A field hidden from you is stripped out of the snippet before the snippet reaches you.

The consequence is the one that matters: the knowledge base cannot hand you anything you could not have opened directly. It is not a side door around your permissions, and it does not become one as your document pile grows, because there is no separate access list to keep in step. The rules you already wrote are the rules search obeys.

Collections, so a team can keep things to itself

Organisation documents live in named collections, and each collection carries its own access grants. Finance keeps its material to finance, the warehouse keeps its procedures to the warehouse, and the company-wide handbook stays company-wide — without standing up a second tool, a second login and a second set of people to remove when someone leaves.

A second tool is where access lists go stale. A collection is governed the same way as everything else in your workspace.

Ask Axis can answer from it

The knowledge base is one of the tools Ask Axis is allowed to reach for. Ask a question in the assistant and it can ground the reply in your uploaded documents and your indexed records instead of guessing from general training.

The permission rule does not soften when a machine does the searching. The assistant queries under the permissions of the person who asked, so two people can put the same question to it and correctly get different answers. Neither of them sees anything they could not have found by searching themselves.

Shared drives are good at storing files and weak at answering questions: they match the words you typed against the words in the file, so material that says the same thing in different words stays hidden. Wikis are good at authoring and linking, and AxisIQ is not attempting to replace that: there is no page tree here, no comment threads, no co-authoring. This is the retrieval layer; it assumes the useful material already exists in some form.

The honest trade is this. A wiki makes writing pleasant. A knowledge base with permission checks on every result makes finding safe. If your problem is that nobody writes anything down, no search tool fixes it — write things down first, then make them findable.

What it does not do

It finds meaning, not exact strings. Searching for order #4712 is the wrong job for this tool — that is what record search and filters are for, and they are exact where this is approximate. Use the knowledge base for "how do we handle a short shipment", and record search for "which shipment was short".

It depends on an embedding model. Hosted AxisIQ provides one; if it is ever unavailable, the knowledge base reports itself as unavailable rather than silently degrading into a keyword search that pretends to be more. We would rather a feature say it is off than quietly do a worse job under the same name.

It only knows what you have given it. Nothing uploaded and no record types switched on means nothing to find, apart from AxisIQ's own product documentation, which is there from the start.

Where it fits

The knowledge base is internal by nature — it is the layer your own team searches. Publishing to the outside world is a different job, handled by Content Studio, the headless CMS this site itself runs on.

To set it up in order — collections, uploads, the record types worth indexing, and who may query what — follow the knowledge base walkthrough.

FAQ

Can someone find a document they do not have access to? No. Access is evaluated on each result at query time, so a document in a collection you cannot query never appears — not as a hit, not as a title, not as a snippet.

Does it search my orders and customers, or only uploaded files? Both, once someone who manages the knowledge base opts a record type in. Indexing is off by default per object type; switch a type on and its records become searchable by meaning, and they re-index in the background whenever they change.

Is this RAG for my business data? Yes, in the usual sense of the term: your documents and records are indexed so an AI answer can be grounded in them. The difference from a typical setup is that retrieval is filtered by the asking person's permissions, including hidden fields being dropped from snippets.

What happens if semantic search is unavailable? It reports itself as unavailable, and the rest of AxisIQ is unaffected. Matching by meaning depends on an embedding model, which hosted AxisIQ provides; if it cannot be reached there is nothing to search against — your documents and records stay where they are.

Can I keep private notes that nobody else sees? Yes — personal material is tied to your account alone; not even the workspace owner can see it. It is for the working notes you would otherwise leave in a text file on your laptop.