Hey business developers and Retool team,
On the road to implement agentic AI in Retool, I've figured out my path and I thought it could be helpful to share with you.
** I cannot yet add the JSON file here, as I'm a new user of the forum.*
In order to create agents, our AI models need to be able to call tools. Tools are actually nothing more than functions.
**Step 1: **
Go to Retool Workflows and create a function named getBeer (so it can be called using getBeer()). I'm creating a REST call to https://random-data-api.com/api/v2/beers which provides me with a random beer in JSON-format each time it's called.
Step 2:
Create an AI action called createPlan that generates text. Let it create the plan using a smaller model, such as gpt-4o-mini.
Input:
Hi, I'm curious to know about a random beer. I'd like to know its name, its style and its hop, so make sure that's part of the plan.
System Message:
Propose a plan to solve the task. You have access to 1 action but you cannot execute it:
* getBeer() which retrieves details of a random beer.
Temperature can be set to 0.7 to reduce creativity.
Step 3:
Create an AI action that executes the function based on the plan. It needs a model that can execute functions, so I used gpt-4.1.
Input:
The plan you should execute is:
{{ createPlan.data }}
You have access to action getBeer()
System Message:
You will receive a plan. Execute the plan and possible actions and only provide the output of the last step after executing all steps.
Temperature to 0.7 here as well.
Step 4:
Run the workflow and see how it returns the name, style and hop contents of a random beer.
Hope this helps you build agentic AI that can use tools (functions) in Retool Workflows.
Let me know if I can assist with anything.
Robin
