Glossary
Tool Calling (Function Calling)
Tool calling, also called function calling, is a capability that allows a language model to request the execution of a defined external function or API during a conversation, then incorporate the result into its response.
General definition
Tool calling extends a language model beyond text generation by giving it a structured way to interact with the outside world. The developer defines a set of tools, each described by a name, description, and parameter schema. When the model determines it needs information or needs to take an action it cannot handle from its training alone, it emits a structured tool-call request instead of a text reply. The application executes the function and returns the result to the model, which then continues the conversation.
Common uses include querying live databases, fetching current information from APIs, running code, sending messages, reading files, or any action the developer wants the model to be able to trigger. The Model Context Protocol (MCP) is an emerging standard for defining and serving tools to AI agents in a vendor-neutral way.
- Tools are defined declaratively (name, description, JSON Schema parameters); the model chooses which tool to call based on the description
- Parallel tool calling allows modern models to request multiple tool executions in a single turn
- The application layer controls whether a tool call is executed, enabling human-in-the-loop approval for sensitive actions
- Prompt injection attacks can attempt to trigger unintended tool calls, so minimal-permission tool design is important
In the Ethora ecosystem
The Ethora AI SDK exposes tool-calling capabilities so that agents built on the platform can query your product’s own data, look up user records, trigger notifications, or hand off to a human agent within a chat thread. Tools are registered and permissioned at the application level, which means you control exactly what each agent is allowed to do.
When combined with self-hosted LLM agents, tool calling keeps sensitive function execution entirely inside your infrastructure. The model never sends your data to a third-party cloud to invoke a tool; the call happens in your own environment, which is a meaningful compliance advantage in regulated verticals.
