Skip to main content

Tool calling

Tool calling is how a Builder turns a user request into concrete actions. It selects tools and executes them to produce results.

How it works​

  1. Interpret the intent from the user’s message.
  2. Select a tool (Action, Procedure, Variant, Toolbox, or Artifact).
  3. Provide inputs using the tool’s schema.
  4. Execute and return the output in chat.

Composable by design​

Because tools share a common input/output contract, a Builder can chain multiple tools directly or route execution through a Toolbox. This enables workflows like:

  • Finding a document.
  • Transforming it (e.g., Word → PDF).
  • Sending it via email.

Toolbox vs. exposing every tool directly​

Using a Toolbox helps keep a Builder’s context window lean. Instead of exposing a large set of tools directly to the model, you reference them inside a Toolbox and let the agent resolve the right execution path only when needed.

Behind the scenes, a Toolbox lets the agent:

  • Call one specific referenced tool.
  • Combine multiple referenced tools as a command-line-style execution chain.

This reduces context clutter and improves execution efficiency because the Builder does not need every referenced tool definition loaded into the active context window. In practice, this improves:

  • Context efficiency: fewer tool definitions competing for space in the prompt.
  • Token consumption: less overhead from exposing a large tool surface directly.
  • Execution time: the agent can resolve a precise tool or chain without unnecessary back-and-forth.