The Trainings feature in Chatmode allows you to enhance your AI agents with domain-specific knowledge. By providing your own data sources, you can make your agents more knowledgeable about your business, products, services, and policies.

​
Understanding Trainings

Training in Chatmode uses vector embeddings and retrieval-augmented generation (RAG) to improve your agent’s responses:
  1. Your training data is processed and converted into vector embeddings
  2. When a user asks a question, the system searches for relevant information in your training data
  3. The agent uses this retrieved information to generate more accurate and informed responses
This process allows your agent to leverage your specific knowledge while maintaining the natural conversational abilities of the base AI model.

​
Types of Training Data

Chatmode supports the following types of training data sources:

​
Direct Text Input (TEXT)

Manually enter or paste text content directly, such as:
  • Product descriptions
  • FAQs
  • Company policies
  • Troubleshooting guides
  • Content extracted from documents (PDF, Word, etc.) manually or via other tools.

​
Website URL (URL)

Provide a single URL to a specific web page. The system will attempt to extract the main textual content from that page for training.

​
Google Shopping XML (GOOGLE_SHOPPING_XML)

Provide a URL to your Google Shopping XML feed. The system will extract product information from the feed and use it for training. This type is optimized for product catalogs and uses item-level chunking for better indexing.
Currently, direct file uploads (PDF, DOCX) and full website crawling are not supported within the training creation process itself. Content from files needs to be extracted and provided as text.

​
Creating a New Training

To add training data:
  1. Navigate to the Trainings section in the sidebar
  2. Click the New Training button
  3. Give your training a descriptive Name.
  4. Select the Data Source Type (TEXT or URL or GOOGLE_SHOPPING_XML).
  5. Provide the Content (either paste text or enter a URL).
  6. Select the API Key (e.g., OpenAI) to use for generating embeddings for this training data.
  7. Configure processing options (see below).
  8. Click Create Training to start the asynchronous processing.
Trainings are associated with Agents after creation, typically when editing an agent.

​
Training Configuration

When creating a training, you can configure:

​
Content Processing

  • Chunk Length: The approximate maximum size (in characters or tokens) of each piece the content is divided into for embedding. Default: 1000.
  • Chunk Overlap: How much content (in characters or tokens) should overlap between consecutive chunks to maintain context. Default: 200.

​
Managing Trainings

​
Training Status

After initiating a training, you can monitor its status in the training list:
  • PENDING: Training is queued for processing.
  • PROCESSING: Content is being extracted and embeddings are being generated.
  • COMPLETED: Training is fully processed and ready to be associated with agents.
  • FAILED: Processing encountered an error. Check logs or try again.

​
Training Details

View basic information about your trainings in the list:
  • Name
  • Type (TEXT, URL, or GOOGLE_SHOPPING_XML)
  • Status
  • Creation Date

​
Training Best Practices

For optimal results with training:
  1. Focus on Quality: Provide accurate, well-structured information
  2. Be Comprehensive: Cover common questions and scenarios
  3. Update Regularly: Keep your training data current as information changes
  4. Structure Appropriately: Organize content logically with clear headings
  5. Test Thoroughly: Verify improvements with realistic queries after training

​
Content Guidelines

  • Break information into clear sections with descriptive headings
  • Include common questions and their answers
  • Provide specific examples and use cases
  • Use consistent terminology throughout
  • Avoid heavily formatted content when possible

​
Example Training Workflows

​
Product Knowledge Base

  1. Upload product manuals, specifications, and FAQs
  2. Add website URLs for product pages
  3. Include common customer questions and appropriate answers
  4. Train your agent to handle product inquiries

​
Internal Policies

  1. Upload company policy documents
  2. Add procedure guides and employee handbooks
  3. Include common HR questions and appropriate responses
  4. Train your agent to assist employees with policy questions

​
Evaluating Training Effectiveness

After adding training, evaluate its impact:
  1. Test Queries: Try asking questions that should be answered with your training data
  2. Review Conversations: Examine real user interactions to identify knowledge gaps
  3. Track Metrics (if applicable): Monitor changes in resolution rates and user satisfaction
  4. Compare Responses: Note differences in agent responses before and after training

​
Troubleshooting

Common issues and solutions:
IssuePossible Solutions
Agent not using training dataCheck if training is COMPLETED and associated with the correct agent(s)
Incorrect information in responsesReview the original source content used for the specific training
Slow processingTraining is processed asynchronously. Large inputs may take time.
Training failedCheck the validity of the URL (if used) or the API Key selected. Simplify content and retry.

​
Next Steps

After setting up trainings:
  1. Test your agent with questions related to your training content
  2. Monitor conversations to evaluate training effectiveness
  3. Update your agent’s prompt to leverage the training data effectively