We are currently experiencing an issue with the AI Assistant module on our VitalPBX system, and we are hoping the community or support team can help clarify or resolve this.
Issue Description:
After the end of a conversation handled by the AI Assistant, the system fails to send the summary email to the address specified in the “Notifications E-mail Addresses” field under the AI Assistant settings. This is happening despite the email address being explicitly included in the “Assistant Instructions”, following the guidelines outlined in the official VitalPBX documentation.
We have verified that our email configuration is working correctly, as other email-based features are functioning without issue. Additionally, we confirmed that call transfers to a live operator are working as expected through the AI Assistant.
System Details:
VitalPBX Version: 4.5.0-7
AI Assistant Version: 4.5.0-5
Is this a known issue with this version of the AI Assistant, or is there an additional step we might have missed in the configuration? Any guidance or suggestions would be greatly appreciated.
After extensive testing, I would like to provide further insight into the issue regarding the AI Assistant not reliably sending the conversation summary email at the end of a call.
New Observations:
The behavior of the summary email sending process appears to be erratic and inconsistent. It seems to depend heavily on how the Assistant Instructions are interpreted by the system. Even when using clear and explicit directives, such as:
“Always send an email at the end of the call or when the user hangs up.”
…the assistant sometimes sends the email, and other times it does not, likely depending on the context of the interaction or how the instruction is semantically processed.
Concerns Identified:
Unreliable Summary Email Delivery: This undermines the ability to consistently monitor or follow up on user interactions through the configured “Notifications E-mail Addresses”.
Lack of Logging or Transcription: There are currently no accessible logs or conversation transcripts within the VitalPBX interface related to the AI Assistant’s operations. This makes it difficult to audit, troubleshoot, or improve the assistant’s behavior based on past conversations.
Suggestion for Future Development:
To enhance the reliability and maintainability of the AI Assistant, I recommend considering the integration of:
Detailed logging for each session.
Transcripts stored per interaction.
Dashboard access to session history and diagnostics.
Improved handling of critical instructions (e.g., guaranteed post-call actions).
These additions would provide much-needed transparency and allow administrators to better understand and optimize the assistant’s performance through real-world context and feedback loops.
Looking forward to hearing if others have experienced similar inconsistencies, and whether there is any roadmap or workaround for this.