It is useful for scenarios where the Agent needs to perform calculations, transform data, validate inputs, or execute custom business logic that goes beyond standard Tools.
🔹 How It Works
-
You configure the Tool with a description that explains in plain language what the code does.
The Agent decides to call this Tool based on that description.
-
You can define properties (variables) that the Agent will collect from the user before running the code.
- Each property has a name, type, and description.
- The description tells the Agent how to extract and format the data from the conversation.
-
The Tool contains a code block, where you implement the function
main().- The Agent will pass the collected properties as inputs.
- The function should return the processed output.
- Important: If the Agent decides to use the Tool but some properties are missing, it will first ask the user for those values, then execute the code.
🔹 Configuration Fields
When creating a Code Execution Tool:- Name → A clear label, e.g.
Discount Calculator - Description → Natural language explanation, e.g.
“Executes a calculation to determine the final price after discount.” - Properties → Variables collected from the user, e.g.:
price(number) → “The original product price in USD.”discount(number) → “The discount percentage to apply.”
- Code → A JavaScript function inside
main()that processes the input and returns the result.
🔹 Example Setup
Name: Discount CalculatorDescription: “Executes a calculation to return the final product price after applying a discount.” Properties:
price(number) → “The original price in USD.”discount(number) → “Discount percentage (0–100).”
🔹 Example Use Case
User: “The product costs 200 dollars, and I have a 15% discount. What’s the final price?”- Agent decides to use the Discount Calculator Tool.
- Properties collected:
price = 200,discount = 15. - Code executes:
200 - (200 * 0.15) = 170. - Agent responds: “The final price after discount is 170 USD.”
🔹 Best Practices
- Keep the description simple → this ensures the Agent knows when to use the Tool.
- Validate inputs inside your code (e.g., check if numbers are positive).
- Return structured outputs in JSON, so the Agent can use the result consistently.
- Test internally before exposing the Tool to users.
✅ Key Takeaway
The Code Execution Tool gives your Agent flexibility to handle custom logic.With clear property descriptions and well-written code, your Agent can dynamically collect inputs, process them, and deliver accurate results in real time.