This feature would allow users to set up a list of "blocked terms" or patterns. Before a message is sent to the LLM, input validation would be performed to check if any of the blocked terms or patterns appear within the draft message.
This could be further extended to enable automatic input sanitisation for certain terms, e.g. the user may wish to substitute mentions of their name (or organisation's name) with some other text, or else completely block the draft message from being able to be submitted.
This would help mitigate the user unintentionally submitting confidential or identifying info during an absent minded copy/paste.
This would be optional config, so should not affect existing workflows or user experience.
I imagine it as being an optional form field in the settings (possibly multiple fields for "block" or "substitute"), or it could be something like an expandable list type of form field where each row has a text input + dropdown determines what type of input validation will be performed.
