Training Overview
Learn how to train your chatbot with your own content so it can answer questions accurately.
How Training Works
When you add content to your chatbot, Leezy:
- Processes the content (text extraction, chunking)
- Creates embeddings (converts text into AI-readable format)
- Stores the knowledge in a vector database
- Retrieves relevant information when users ask questions
This approach is called RAG (Retrieval-Augmented Generation). Your chatbot retrieves relevant content from its training data before generating a response.
Data Sources
You can train your chatbot with multiple types of content:
| Source Type | Description | Best For |
|---|---|---|
| Documents | PDF, DOCX, RTF, TXT files | Manuals, policies, reports |
| Websites | Crawled web pages | FAQ pages, help centres, product pages |
| Q&A Pairs | Question + Answer | Specific questions, corrections |
| Notion | Connected workspace pages | Internal wikis, runbooks, playbooks |
:::tip Combine Sources Use multiple source types for comprehensive training. Documents provide depth, websites give structure, and Q&A pairs handle specific cases. :::
Accessing Sources
- Go to your dashboard
- Select a chatbot
- Click Sources
Training Best Practices
Quality Over Quantity
- Use accurate, up-to-date content
- Remove duplicate or conflicting information
- Focus on content users actually ask about
Structure Content Well
- Use clear headings and sections
- Keep paragraphs focused on one topic
- Include FAQs when possible
Test Frequently
- Use the Playground after adding content
- Ask questions your users would ask
- Verify responses are accurate
Document Processing
When you upload content, you'll see processing status:
| Status | Meaning |
|---|---|
| Pending | Queued for processing |
| Processing | Being analyzed and embedded |
| Ready | Successfully processed, active |
| Failed | Error during processing |
Processing time depends on content size. Most documents complete within a few minutes.
Managing Sources
View All Sources
The Sources page shows all training data with:
- Source type and name
- Processing status
- Character count
- Date added
Delete Sources
- Select the source(s) to delete
- Click Delete
- Confirm deletion
:::warning Removing Knowledge Deleting a source removes that knowledge from your chatbot immediately. :::
Storage Limits
Training data storage is organization-wide and depends on your plan:
| Plan | Storage Limit |
|---|---|
| Free | 500 KB |
| Starter | 50 MB |
| Pro | 50 MB |
| Business | 50 MB |
:::info Free Tier Storage 500 KB is enough for a handful of text-heavy pages or one short PDF. If uploads start failing on the free tier, storage — not file size — is usually the limit you hit. :::