Sensitive information in agent logs

Agents often interact with highly sensitive data during live calls, including Personally Identifiable Information (PII) (e.g., social security numbers, home addresses) and Financial Details (e.g., credit card numbers, bank routing info).

To prevent this data from being stored in plain text within your agent logs, you must configure your logging levels correctly.

To configure logging levels:

1. Navigate to AI Agents > Agents.
2. Locate your agent, and click Edit.
3. In ‘Miscellaneous’ tab, configure the ‘Logging’ parameter.

The following logging levels are supported:

• enabled – enables logging of all events, including user utterances and LLM responses.
• disabled – disables logging of all events, except for start, pick-up, prompt, summary, and task_switch, and ensures that remaining events contain no sensitive information.
• masked – masks content in all log events - i.e. you will still see that there was user utterance or LLM response, but the content will be masked.

In multi-agent topologies, logging should be independently configured for each agent that participates in the conversation. For example, you may disable or mask logs for specific sub-agents that handle sensitive conversation parts, while keeping logs enabled for the rest of the agents.

For Flows, logging can be configured for each conversation node.

When agent logging is disabled or masked, corresponding user utterances and LLM responses are automatically marked as “containing sensitive information”, thus preventing them from being logged in call transcript and LiveHub platform logs. For a detailed description, see VoiceAI Connect > Bot integration > Basic behavior > Logging and privacy > Masking sensitive information from bot.