Knowledge bases

A knowledge base is a named collection of documents belonging to your organization. Put a document in one, and two things become possible: you can search it directly, and Klu can retrieve from it and cite it when answering.

Everything about knowledge bases lives under Knowledge in the Hub sidebar, with two entries: Documents (everything you have uploaded) and Knowledge bases (the collections themselves).

Why collections at all?

A single organization can keep several knowledge bases and it usually should. The reason is retrieval quality: a base that mixes model contracts, internal policy and client correspondence answers every question a little worse than three bases that each hold one kind of thing. Name them for what they contain — the create form suggests something like “Legal Documents” — and the search page lets you point a question at one of them.

Creating a knowledge base

Go to Knowledge → Knowledge bases. The top of the page shows how much you have — knowledge bases, total documents — and below it Create new knowledge base with two fields:

Name — required.
Description — optional; what this knowledge base contains.

Press Create base. That’s the whole form. Everything technical about how documents get indexed is chosen for you.

A new knowledge base is not yet read by any assistant. Before Klu can answer from it, give it to an assistant — see Which assistants read a knowledge base.

Each row in the list shows the name, an Active or Inactive pill, the description, and a document count.

Editing, deactivating and deleting

Open a knowledge base to reach Edit Knowledge Base — change the Name or Description, or flip the Active switch, then Save changes.

Deactivating is the gentler alternative to deleting. An inactive base stays in this list with all its documents, but drops out of the pickers: you cannot upload to it and it is not offered for search.

Deleting is under Danger Zone on the same page, and it is permanent — the confirmation tells you how many documents will go with it. There is no undo.

Folders

A knowledge base’s page lists its contents as folders and documents, with the path you are in shown above them. Press New folder, enter a Folder name and press Create folder to add one where you are. Upload on that page adds documents to the folder you have open, and each document’s page shows the Folder it sits in.

Which assistants read a knowledge base

A knowledge base is read only by the assistants it has been given. The Available to panel at the top of a knowledge base’s page names them.

Add assistants opens a list of your organization’s assistants; tick the ones that should read this base and press Add. The × beside a name removes it.
Available to all assistants is a switch that opens the base to every assistant in your organization, including ones created later. Turning it off again leaves the assistants you added individually in place.

A base nobody may read yet says No assistant can read this knowledge base yet. A base switched to Inactive is read by no assistant, whatever this panel says.

Changing this needs the Manage AI assistants permission; without it you see the panel but not its controls. The reference collections your product includes (see below) are not part of this — every assistant reads them.

Uploading documents

Go to Knowledge → Documents and press Upload, or press Upload on a knowledge base’s page to have that base, and the folder you are in, pre-selected.

Knowledge Base — select which base to add the documents to. Only active bases are offered.
Documents — drag files onto the drop zone, or click browse. You can select several at once. To add a whole folder, press Upload folder instead: its subfolders are kept, and each file is uploaded on its own.
Press Upload & Process.

Each file then gets its own row showing how far it has got — Uploading…, Processing, Ready — and, if it was refused, why. When every file has gone through you are taken to the knowledge base. If some failed you stay on the page: press Retry on a row, or Retry failed files for all of them, or Go to the knowledge base.

Supported files

PDF, Word, Excel, PowerPoint, images and text files — specifically .pdf, .docx, .xlsx, .xlsm, .pptx, .png, .jpg, .jpeg, .webp, .gif, .tiff, .tif, .bmp, .txt, .md and .csv. Images are read with text recognition, so a scanned page works.

Uploading the same file twice into the same base does nothing the second time — its row reads Skipped, with Already in this knowledge base.

While documents are processing

On the knowledge base’s page and in the Documents list, each document carries a status:

Status	Meaning
Pending	Queued, not started yet.
(progress badge)	Being processed. The badge names the step in progress, and for a PDF also the page it has reached.
Ready	Searchable and available to Klu.
Failed	Something went wrong; nothing was indexed.

The list refreshes itself every few seconds while anything is still in progress, so you can watch a batch complete without reloading. Filter tabs — All, Ready, Processing, Failed — each carry a count.

A long document takes longer than a short one, and a scanned image takes longer than text. There is nothing to do but wait; nothing is lost if you navigate away.

When a document fails

Open the document. The reason is shown in full at the top of its page — most often an unsupported file type or a file over the size limit. Fix the cause and upload the corrected file, or press Reprocess to try the same file again (worth doing if the failure looked transient).

Working with a document

The Documents list is searchable by filename or title, sortable (Newest first, Oldest first, Filename A-Z), and filterable by knowledge base and by where the document came from.

Open any document to see:

Its page, word and section counts, and the language detected.
Extracted Text — the text kOS actually read out of your file, with Expand to see more. This is the page to check when an answer seems to have missed something obvious: if the text is wrong or empty, the file was unreadable, not the assistant.
Details — which knowledge base and folder it is in, its source, format, when it was uploaded and by whom.
Actions — View Original to open the file you uploaded, Reprocess, and Delete.

Deleting a document removes it and its index entries immediately, with no undo.

How documents are used in answers

There are two ways to get at your documents.

Klu retrieves automatically

When you ask Klu something answerable from your documents, it searches on its own — you do not pick a knowledge base first, and there is no picker in chat. It searches every active knowledge base that the assistant you are talking to may read (see Which assistants read a knowledge base), plus any curated corpora your product includes (see below). If Klu seems not to know a document you uploaded, check that its knowledge base is available to that assistant.

Answers arrive with their sources attached. Each source gives the document name, the page it came from, the section heading where applicable, and the passage used, with a link to open or download the original. Read more in The Klu assistant.

One behaviour worth understanding: when you attach a file to the message, Klu answers from that file’s text rather than searching your knowledge bases. That is deliberate — it stops a question about the contract in your hand being answered from an unrelated file in your library. If you want the question answered against your knowledge bases as well, say so in the question.

Searching directly

The search page — open it from the Search Knowledge Base tile on Home — asks for a Query and one Knowledge Base, and returns up to ten passages. Each result shows the document title and filename, the page, the section heading, a highlighted excerpt and surrounding context, plus a link to the source or a Download.

Use this when you want to see the passages themselves rather than an answer built from them — checking whether something is in your library at all, or verifying a quote.

Curated corpora that are not yours to manage

Some products include reference collections curated by KlusAI rather than uploaded by you. Lex, for example, includes a Romanian legislation corpus covering the civil, civil-procedure, criminal, criminal-procedure, labour and fiscal codes.

These behave differently from your own knowledge bases on purpose:

They do not appear under Knowledge bases, and their documents do not appear in your Documents list. You cannot upload to them, rename them or delete them.
They are offered on the Search page, and Klu retrieves from them automatically.
Their citations carry extras your own documents cannot have — a link to the official source, and a stamp saying which consolidated version the text reflects.

Which corpora are included depends on your product. See Lex and Med for what each one brings.

Who can see what

Knowledge bases belong to the organization, not to a person. Any colleague with knowledge access sees the same bases and the same library documents — that is the point of them.

There is one exception, and it works the other way: documents that arrive through a conversation with Klu rather than through the upload page are visible only to the person who sent them. Those never become shared library content unless they were uploaded through the knowledge base’s Upload page.

There is no per-document or per-base sharing setting. If a document should not be visible to the whole organization, do not put it in a knowledge base.

Limits

There is no cap on the number of knowledge bases, documents, or pages you may keep. The one hard limit is 100 MB per file.

The numbers you see on these pages — documents, pages, words, indexed sections — are counters showing what you have, not allowances you are spending.