Tool Calling
Test how your LLM interacts with tools before building real integrations
Adaline’s Playground offers a safe environment to test how your LLM interacts with tools without making API calls. This feature lets you develop and refine tool integrations before you implement it in production evironment.
Playground Tool Call Workflow
When you run a prompt that causes the LLM to call an external tool, the Playground displays the tool call in a special format but doesn’t execute it.
The image above shows a tool defined in the editor.
The image above shows the Tool Calling response after the prompt has been executed in the Playground.
This gives you visibility into what the LLM would request while maintaining complete control over the process. Through this method, Adaline does not fetch any data from your API. Instead, Adaline reflects the behaviour of the LLM with Tool-Calling capabilities.
How Tool Calls Work in the Playground
When the LLM determines it needs external data, it generates a tool call request that appears as a distinct message in the conversation flow. The tool response is given via the Assistant message.
The tool call shows:
- Function name being requested
- Required parameters and their values
- “Add response” tab to manually input the tool response Essentially, the LLM is seeking an API response.
Since we don’t support API calls yet, you can manually plug in the response and complete the tool call. This will let the LLM think that the API call has been made and the results have been fetched from the API.
Once the response is received, the LLM will generate the final result via the Assistant role message.
Adding Tool Calling Response
Follow these steps to manually implement the Tool Calling response:
Execute Your Prompt
Run your prompt in the Playground as you normally would.
Identify Tool Call Triggers
Monitor the conversation for tool call messages. These appear when your prompt logic determines that external data is needed to complete the user’s request.
Simulate Tool Response
Click the “Add Response” button to provide simulated data:
- Enter response data in the format your tool would return.
- Use realistic sample data that represents actual tool outputs.
Execute the prompt to completion
Click on the Run button to get the final output from the LLM.
To learn how to define a tool, check out Tools.