A SaaS tool that automates sanitization and filtering of AI model responses for webhook channels and others, ensuring no internal data or reasoning leaks.

The webhook channel (and others that do not implement draft updates) POST the model's full unstripped response back to its callback URL. When the model encodes reasoning as inline <think>...</think> tags, these tags appear verbatim in the outbound webhook body. The sanitize_channel_response function does not strip these tags, although a helper strip_think_tags_inline exists and is used only in draft update paths. This causes end users or callback consumers to receive internal chain-of-thought reasoning that should be hidden. The proposed fix is to modify sanitize_channel_response to also strip these tags and add unit and integration tests to ensure no leaks.