Skip to main content
Code Tools let you define custom logic using JavaScript or Python that agents can invoke as part of a workflow. They bridge the gap between LLM reasoning and deterministic, programmatic operations, enabling your agents to reliably transform data, call APIs, or run business logic.

​
Step 1: Basic Info

Provide the basic details of the tool. Name: Unique identifier for the tool. Description: Briefly describe the tasks of the tool. This description is used by the LLM while selecting the tools for task execution.

​
Step 2: Code and Config

Runtime: Select the execution environment for your tool’s code. Agent Platform supports JavaScript and Python. Code: Write the business logic for processing.
  • You must define a main() function. This is the entry point the platform calls when the tool is invoked.
  • The main() function must return a value. Example: return main($param).
  • Parameters are injected into the code using a prefix.Forexample,aparameternamedcustomernameisaccessedas‘ prefix. For example, a parameter named customer_name is accessed as `customer_name` in your code.
Parameters: Define the inputs this tool accepts. These parameters are sent to the LLM as a JSON Schema. The parameters are injected into the code using the $ prefix. Click + Add to add a new parameter manually, or click Parse to auto-detect parameters from your main() function signature.
FieldDescription
NameThe parameter name, prefixed with (e.g.,(e.g.,data). This is how you reference it in code.
TypeData type - String, Number, Boolean, Object.
RequiredToggle whether this parameter must always be provided.
DescriptionA natural-language description sent to the LLM to explain what this parameter represents (e.g., “The customer’s full name”). This field is mandatory.
Default valueAn optional fallback value that is used if the parameter is not supplied.
Generated JSON Schema The platform automatically generates a JSON Schema from your defined parameters. You can inspect it by expanding View Generated Input Schema. This is sent to the LLM to describe the required parameters for the tool execution. Advanced Settings
  • Return Type: Specifies the data type of the value your tool returns. This is used in the DSL signature and the LLM schema. Common values: object, string, array, and boolean. Default: object
  • Memory (MB): The memory allocated to the tool’s isolated container during execution, in megabytes. Default: 256
  • Timeout (ms): The maximum time (in milliseconds) the tool is allowed to run before it is terminated. Set this based on the expected complexity and latency of your tool’s logic. Default: 5000.
  • Variable Namespace: Namespace that the tool can access. By default, every tool can access ‘Default’ namespace.

​
Step 3: Test & Review

Provide sample values for the input variables and Run Test. Verify the response from the tool. Check Logs and parameters for debugging purposes.