Does Retell forward the system prompt verbatim to OpenAI (for prompt-cache eligibility)?

Support Help
1 
Running a single-prompt agent on gpt-4.1 (~8k-token general_prompt,
~17 turns/call). Looking at OpenAI prompt caching for a 50% discount
on turns 2+, but it only fires if the prefix is byte-identical
across requests.

Questions:

1. Does Retell forward `general_prompt` verbatim, or inject dynamic
   content (timestamp, call_id, variables) into the prefix?

2. If there's injection, can I structure my prompt so the static
   portion stays contiguous and dynamic bits go at the end?

3. Is OpenAI prompt caching active on Retell-managed gpt-4.1, or
   opaque to customers?

Already saw thread #2610 and the 3,500-token flat-rate changelog —
trying to figure out if caching is a viable lever before migrating
to Conversation Flow.

Thanks!
2

Hey @nbomb Thanks for the details, I have sent your questions to the team for a detailed answer.

Best Regards

3

Hello @nbomb

  1. We forward it verbatim but inline dynamic variables where they are referenced. So if you reference dynamic variables in your general prompt and they change during a call, that could break the prompt caching. But if the dynamic variables don’t change or if they aren’t referenced, the caching will remain intact.

  2. You can reference the changing dynamic variables at the end of the prompt.

  3. Yes, it is enabled.

Thank You