There's broadly two types of GHL users:
  1. SaaS: Sell GHL features that work out of box.
[2]. Agencies/SWAS: Sell add-on services. Requires customization specific to the end user/customer's requirements.
Guardrails are used to:
  1. Prevent prompt & code injection
  2. Follow the necessary compliance & regulations: Protect Customer Info (PII), GDPR, HIPAA
  3. Avoid certain types of content: Offensive, political etc.
  4. Hallucinations
[5]. Real world application instructions and more.
This request is specifically to help Agency [2] users that make use of advanced prompts and need the customizability only available with the text based prompt editor. This gives Agency owners granular control over the exact order of instructions to be followed by AI.
SaaS owners can stick to Basic mode, refer image below.
Issues right now:
  1. Voice AI asks for email every time an appointment in booked even when the prompt instructions say otherwise. I'm assuming this is hardcoded prompt guardrail inside GHL. When calling my barber I would never give them my email, I'd rather just give him my number and interact with him this way.