The end-of-conversation summary in Agent AI is generated using the Kore.ai XO GPT module.
If clients prefer to use a different model (for reasons such as language or quality), they can host their model in GALE or another server and provide their API endpoint. The custom summarization workflow uses the “Custom Dialog Task” option – when the conversation ends, Agent AI will trigger the configured dialog, which will fetch the conversation, generate the summary, and display or save it as needed.
Steps to configure auto summarization with custom dialog tasks
- Set up a service node to get the conversation transcript using the ChatHistory API; the API includes the conversation summary.
- Send this transcript to the Summarization endpoint, and save the result in a Context Variable.
- Create a string-based entity node to display the summary from the Context Variable.
- Ensure the dialog is enabled for the Agent AI chat/voice channel and the bot is published.
- Go to Agent AI > Widget Settings > Agent AI Settings > Agent AI Channel Specific Settings > Auto Summarization > Auto Summarization Model, and select a dialog task under the “Custom Dialog Task” option.
Dialog Task Flow for Summarization activities: