Tool Call

Tool Call is the programmatic mechanism where a large language model decides to execute a registered function, API request, or database query rather than generating standard text. It allows models to interact with external systems.

How it Works

A tool call bridges the gap between text generation and code execution.

• Function Registration: The orchestrator presents the LLM with a list of available tools, including their names, descriptions, and expected parameters in a JSON schema.
• Model Output: When the model requires external data, it outputs a structured payload (usually JSON) containing the tool name and argument values (e.g., execute_sql(query="SELECT * FROM table")).
• Orchestrator Execution: The application parses the model’s output, runs the corresponding function, and returns the output to the model context.

Lakehouse & Agentic Relevance

In a data lakehouse, AI agents use tool calls to perform operations on database tables. For instance, the agent makes a tool call to run SQL queries or search metadata. Dremio acts as a primary tool endpoint for these calls. By exposing a high-speed JDBC/ODBC or Arrow Flight interface, Dremio allows orchestrators to execute tool calls on behalf of the agent, delivering query results back to the model within milliseconds.