Prompt Recipe¶

A Prompt Recipe applies a prompt to each row of an input dataset and writes the results to an output dataset in the Flow.

The prompt can use a standard LLM, a Retrieval-Augmented LLM, or an Agent.

Use Prompt Studios to design and test a prompt interactively, then use a Prompt Recipe to run it in batch on a dataset.

When to use a Prompt Recipe¶

Use a Prompt Recipe when you want to:

Run the same prompt on many rows of a dataset
Turn work from Prompt Studio into a repeatable Flow step
Use LLM Mesh capabilities such as Retrieval-Augmented LLMs, Agents, Guardrails, and tracing in a dataset-processing workflow

Prompt Recipes in the LLM Mesh ecosystem¶

Prompt Recipes are one of the main ways to run LLM Mesh capabilities in DSS as part of a Flow:

Design and test prompts in Prompt Studio
Use Adding Knowledge to LLMs when the prompt needs retrieval-augmented generation
Use AI Agents when each row should be processed by an Agent
Use Guardrails to apply safety and policy checks to inputs and outputs
Use Tracing to inspect raw traces for analysis and debugging

If your Prompt Recipe relies on an Agent or a Retrieval-Augmented LLM that uses structured request metadata, you can provide additional request context either as a fixed value or from a column in the Prompt Recipe input dataset.