Prompt library

The prompt library lets you store reusable prompt texts as named entities. Prompts can be referenced by ID from multiple plugins, guardrails, and other features, providing a single source of truth for your prompt texts.

Entity structure

{
  "id": "prompt_xxx",
  "name": "English translator",
  "description": "A system prompt for translating content to English",
  "prompt": "You are a professional translator. Translate the following content to English. Return only the translation, no explanation."
}

Field	Type	Description
id	string	Auto-generated with prompt_ prefix
name	string	A descriptive name for the prompt
description	string	What this prompt is used for
prompt	string	The actual prompt text

Where prompts can be referenced

Prompts can be referenced via their ID (prompt_ref) from:

• Request body modifier — system prompt for transforming request bodies
• Response body modifier — system prompt for transforming response bodies
• Response generator — prompt for generating responses
• LLM Response endpoint — prompt with Expression Language
• Context validator — system prompt for validating requests
• WebSocket validator — system prompt for validating WebSocket messages
• LLM guardrails — validation prompt for custom LLM-based guardrails

Testing prompts

The admin UI includes a built-in prompt tester. Select a provider and test your prompt directly from the Otoroshi back-office without creating a route.

Admin API

ai-gateway.extensions.cloud-apim.com/v1/prompts

Standard CRUD operations are supported.