Data Sources & Training | Document

Introduction

Document training allows you to train your AI agent using files such as PDFs, Word documents, or other reference materials.

This is useful when important information lives outside your website—for example, internal manuals, policy documents, SOPs, product specs, onboarding guides, or legal documents.

By training on documents, your AI agent can give accurate answers based on your actual source files.

What Document Training Does

When you upload a document:

• Exei reads and processes the content

• The information is converted into structured knowledge

• The agent uses this data to answer relevant questions

Document training helps the agent:

• Respond accurately to detailed or policy-driven queries

• Reference information that may not be public

• Reduce manual copying of long text into the system

Where to Add Documents

To train your agent using documents:

1. Open your AI agent from the My Agents page

2. Go to Knowledge Base

3. Select Document

This section shows all documents currently used to train the agent.

Supported Document Types

You can upload commonly used document formats, including:

• PDF

Make sure documents are readable and not password-protected.

Uploading a Document

To add a document:

• Click Upload Document

• Select the file from your device

• Confirm the upload

Once uploaded, Exei starts processing the document automatically.

You can upload multiple documents, and each one is handled independently.

What Kind of Documents Should You Upload?

Document training works best for structured, reference-style content.

Common examples include:

• Terms and conditions

• Return, refund, and shipping policies

• Product manuals or specifications

• Internal SOPs or playbooks

• HR or IT policy documents

• Compliance or legal documents

• Onboarding guides

If users frequently ask about something that exists in a file, document training is a good fit.

How Documents Are Processed

When a document is uploaded:

• Exei extracts readable text from the file

• Content is broken into logical sections

• The agent uses relevant sections to answer questions

Scanned documents or images without readable text may not train correctly.

Document Status and Management

• Upload new documents at any time

• Remove documents that are outdated

• Replace documents with updated versions

When a document is removed, the agent no longer uses that information.

Keeping Documents Up to Date

If the content in a document changes:

• Upload the updated version

• Remove the old version if it’s no longer relevant

Keeping documents current ensures the agent does not give outdated or incorrect answers.

Combining Document Training with Other Data Sources

Document training works best when combined with:

• Website training (for public-facing content)

• Text training (for specific instructions or exceptions)

• Catalog training (for Shopping Assistants)

Each data source fills a different gap and improves overall accuracy.

Best Practices for Document Training

• Upload clear, well-structured documents

• Avoid uploading duplicate or overlapping files

• Remove outdated documents regularly

• Use document training for detailed or policy-heavy content

• Test responses after uploading new documents

Well-maintained documents lead to better answers.

Common Mistakes to Avoid

• Uploading scanned or image-only PDFs

• Training on draft or outdated documents

• Uploading too many documents without testing

• Expecting the agent to answer beyond what’s in the files

• If the information isn’t in the document, the agent won’t know it.

How to Validate Document Training

After a document becomes active:

• Open Test AI Agent

• Ask questions based on the document content

• Check for accuracy and clarity

• Use Instant Retrain if corrections are needed

Testing helps confirm the document is being used correctly.

What’s Next?

Now that your agent is trained using documents, you can add more data sources or move toward deployment.

Next User Guides:

Data Sources & Training | Text