Tools

What are Tools

At its core, an LLM can only generate text based on the text it is fed. Therefore, we need a mechanism that allows the LLM to interact with the outside world; this mechanism is called tool invocation.

Specifically, the LLM generates a special string of characters to indicate that it wants to invoke a tool; the LLM service and Optic then run the tool and return the results of its execution to the LLM. You don't need to worry about this process.

For example, when an LLM wants to read a document on your computer, it can invoke a document-reading tool (if available), and that tool will return the document's content to the LLM.

Capability

This means that an LLM must be capable of invoking tools in order to function as an agent. Currently, the vast majority of LLMs possess this capability.

Please refer to this link. If you add a model that Optic is not familiar with, you need to edit it to inform Optic that it has the ability to invoke tools, thereby enabling it to function as an agent.

So, how exactly do you configure tools?

Optic provides some built-in tools, which we'll cover later. These tools (such as web search) are essential features we believe an agent should have, so you don't need to “install” them—simply tap the button in the row above the text input box on the main interface to enable them.

However, for more specific scenarios—such as connecting to an app like Notion—you'll need to use the MCP or CLI. We'll discuss these next.