Has anyone built an n8n workflow for testing prompts?

Questions
1

I’m on a journey to design a complex prompt-testing workflow in n8n, and as a beginner, I’m looking for guidance to bring my idea to life.

Here’s what I’m aiming to build:
I want to create an n8n workflow that processes prompts one at a time. Each prompt should be dynamically combined with a predefined payload during the workflow. The system should then process the combined input, generate results, and provide a detailed comparison of those results to show how each prompt influences the final outcome.

To make this workflow more flexible and scalable, I’d like to integrate a database to manage both prompts and payloads:

  • Prompts should be saved in the database and organized by category, allowing me to easily retrieve specific ones based on their type.
  • Payloads should also be saved in the database and linked to their corresponding prompt categories so they can be dynamically paired during processing.
  • The setup should make it easy to iterate and update both prompts and payloads as I refine my workflow.

This system will help me systematically test and analyze prompt patterns while keeping everything organized and reusable.

As I’m just getting started, I’d love any advice on how to structure this workflow in n8n, including tips or examples from anyone who has tackled something similar.

I’m excited to learn more and refine this workflow as part of my journey into n8n. Thanks in advance for your help!

2

It looks like your topic is missing some important information. Could you provide the following if applicable.

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
5

Hi @Steven_Sexton

You can use on of our database nodes to store the prompts and payloads by category (e.g. Airtable, GoogleSheets or SQL databases like Postgres or MySQL, depending on your preference).
You can use a Set Node to combine different prompts / payloads for processing via your AI of choice.
For the comparison of the results you could capture the responses in either the same database or another (utilizing yet again any of our database nodes).

Why don’t you just get started with building a workflow and if you’re stuck or have specific questions, feel free to post here :slight_smile:

Btw, we’re also working on a workflow evaluation feature, which lets you test and evaluate workflow runs (including performance of a prompt) :wink: Stay tuned!

6

I tried using Postgres to store prompts and responses but hit several issues:

  1. Schema design: Struggled with effectively categorizing prompts and responses while maintaining flexibility
  2. Database setup: Had trouble with proper indexing for performant querying
  3. Response formatting: JSON storage of AI responses became unwieldy

I attempted using GoogleSheets as an alternative but ran into rate limiting and data structure limitations.

Would appreciate guidance on:

  • Recommended schema design for prompt/response storage
  • Best practices for database node configuration
  • Workflow patterns for prompt testing and evaluation

Looking forward to the upcoming workflow evaluation feature!

7

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.