Extending Models with Tools
Models can be extended via Tools, which can be configured to run via an API. Tools are supported for the following model providers:
- OpenAI
- Azure OpenAI
- Anthropic
- AWS Bedrock
To help you understand how we can use tools effectively, we created a short tutorial to show you how they can be used.
Configuring a Tool
When configuring a tool, you need to provide:
- Tool Description or Purpose
- Example Use Cases
- API Call
- Parameters for your API
Example: Setting Up a Google Search Tool
This example shows how to set up a tool to perform a Google Search via an API.
Steps to Configure
-
Navigate to a Project
Select "Tools" on the upper navigation bar, which will allow you to view the tools that are already in your Project.
-
Select New Tool
Select the option to create a New Tool:
This will take you to the library of new tools that you can add, or for you to create a new one.
-
Add a tool to the library
You can either add a tool that is already available to you to the library, or create a new one with a custom definition.
-
Adding a custom tool
When adding a custom Tool, you will see a page that allows you to configure the tool manually.
-
Adding an existing tool"
You can also add an existing tool from the library, which will present you with a
-
Provide authentication
Each Tool has it's own authentication, as dictated by the API Provider. Add this to the Tool by selecting on the corresponding field, and configuring it with the credendial.
As an example, for a tool that requires Bearer Authentication, you would select Bearer authentication, and add the corresponding credential to it. You can select from one of the existing values, or create a new one before adding it.
-
Create an Agent
Add one of the models that support tools to the Agent:
- Edit the Tool
Click the Model block to open the sidebar, and add a tool from the list and save:
- Save the Agent
The tool should appear as part of the model footer:
Testing the Tool
After adding the tool to an Agent, it can be tested via the playground. Test the tool by sending a query that prompts the LLM to use the tool. Tools work with both Streaming and Non-streaming methods in the Playground.
By following these steps, you can extend models with tools effectively and efficiently. If you have any queries, please refer to the respective tool documentation or reach out for support.