Auto-cache: OpenAI prompt key

Derive a deterministic prompt_cache_key so repeated prompts hit the OpenAI prompt cache.

What it does

This transform computes a deterministic prompt_cache_key from the request content and attaches it, together with a retention hint, for OpenAI Responses and Chat Completions upstreams. Repeated prompts with the same prefix then hit the provider prompt cache.

The transform never overwrites a prompt_cache_key that the client already set.

Reference

Property	Value
Type ID	`cache_openai_prompt`
Phase	`request`
Scopes	Provider, Global, API key

When to use

The upstream is OpenAI-compatible and prices cached input tokens lower.
Clients do not set prompt_cache_key themselves.

Configuration

Option	Type	Required	Default	Description
`retention`	string	No	`"24h"`	`24h` or `in_memory`. Cache retention hint sent to the upstream.
`key_prefix`	string	No	`"mzpc"`	Prefix of the generated `prompt_cache_key`.
`key_mode`	string	No	`"prefix"`	`prefix` derives the key from the conversation prefix. `identity` uses one key per identity.
`include_user_in_key`	boolean	No	`false`	Mix the Monoize user into the key so users never share cache entries.
`include_full_input_in_key`	boolean	No	`false`	Hash the full input instead of the stable prefix.

Example rule

{
  "transform": "cache_openai_prompt",
  "enabled": true,
  "phase": "request",
  "models": [
    "gpt-*"
  ],
  "config": {
    "retention": "24h",
    "key_mode": "prefix"
  }
}