Knowledge Base

The Knowledge Base allows you to upload documents that your AI assistant can reference when answering questions, providing accurate and contextual responses grounded in your content.

How It Works

Upload documents to an assistant's knowledge base
Wexio splits them into passages and turns each into a vector with OpenAI's text-embedding-3-large
When the assistant receives a message, the question is embedded the same way and matched against those passages by meaning, not keywords
Matching content is included as context for generating the response
Knowledge base information takes priority over general model knowledge

Because questions and passages are embedded with the same model, the match is cross-lingual: a question in Ukrainian can retrieve a passage written in English, as long as they mean the same thing.

You do not need to translate your knowledge base into every language you support. Write it once, in whichever language it is clearest, and let retrieval bridge the gap. Translating is still worth it for the answer wording - see the assistant's language setting.

Who the AI Shares It With

Each knowledge file, help article and news post can be limited to an audience - who the AI is allowed to use it for.

Audience	The AI uses it for
Everyone	Any visitor. The default.
Signed-in visitors	Only visitors who have authenticated.
A field value	Only visitors whose contact field is one of the values you list - `plan` is `Pro` or `Enterprise`, say.

Audiences are set per locale variant, so a translated article can have its own.

This limits what the AI says, not what the widget shows. A help article stays publicly readable in the widget regardless of its audience - the setting only stops the assistant from drawing on it when answering someone outside that audience.

If a document must not be readable at all, do not publish it as an article.

The assistant knows whether a visitor is signed in and what their plan is, but it never states that back to them. It answers differently; it does not announce why.

Uploading Documents

Go to Settings → AI and select an assistant
Navigate to the Knowledge Base section
Click Upload
Select one or more files
Documents are processed and indexed automatically

Supported Formats

Format	Description
PDF	Documents, manuals, guides
TXT	Plain text files
Markdown	`.md` files with formatting

Answer Quality

Answers are drawn from your knowledge rather than the model's general training, and are checked before they are sent.


Grounded	Every answer comes from your organisation's knowledge, with internal citations back to the source.
Checked	Numbers, prices, dates and links in a reply are verified against those sources before sending.
Honest when it cannot answer	If a claim is not supported, the assistant says it could not find an exact answer and lists the closest topics instead.
Hands over only when it can	It offers a human only when one is actually available.

This is why a wrong price is unlikely but a "I could not find an exact answer" is not a failure - it is the assistant declining to invent one. The fix is usually a gap in the knowledge base, not the model.

Resolution Priority

When answering questions, the assistant checks sources in this order:

Knowledge base (highest priority)
Conversation context
Assistant instructions
General model knowledge (lowest priority)

Managing Documents

Action	Description
View	See all uploaded documents and their status
Delete	Remove a document from the knowledge base
Re-upload	Replace a document with an updated version

Best Practices

Upload comprehensive documentation - Product docs, FAQs, policy documents
Keep documents up-to-date - Remove outdated files and upload new versions
Use clear formatting - Well-structured documents produce better search results
Organize by topic - Upload focused documents rather than one massive file

Knowledge base processing may take a few minutes for large documents. Once processed, the assistant can immediately reference the content.