AI tools
AI tools are specialized functions or capabilities that you can provide to an AI agent. They bridge the gap between the AI’s language processing capabilities and real-world actions or data. By using tools, an agent can look up information in a database, send emails, call external APIs, or execute scripts, making it a dynamic assistant rather than just a text generator.
| The specific instantiation of an LLM using a tool is often referred to as "Tool Calling" or "Function calling". It consists of the LLM generating input parameters for the tool in question and the execution of predefined code acting on those input parameters. |
Example
You have set up an API to an external service, complete with authentication and operations. You want to enable an agent to use this API, in order to perform actions on behalf of the user.
To do so, you define an API tool that points to the API operation that you created. You then edit the agent and select the newly created API tool in the Advanced settings tab.
Your agent can now perform actions against the API operation.
|
The effectiveness of a tool depends a lot on the usefulness of its prompt/description. Think of the description as the explanation for the AI on why/when it should be using a particular tool. Furthermore, some tools depend on the data specified in the original artifact, like property descriptions in the table definition or operation/field descriptions in the API tool. |