Executes all toolCalls concurrently and returns their outcomes in input order.
the tool calls collected for this turn (id + name + raw args)
the tools available for this request (typically request.tools), reduced to the
ExecutableTool shape
Optionaloptions: ToolCallExecutionOptions
optional per-call hook and cancellation token
Executes the tool calls of a single language model response/turn.
Concurrency contract: tool calls emitted within a single model response/turn are executed concurrently. All matched handlers are started before any of them is awaited. A tool handler must therefore not assume that it runs in isolation or in array order, and must be safe to overlap with sibling calls from the same turn. Models serialize dependent calls across turns (by withholding a dependent call until it has seen the prior result), so "issued together in one turn" is the model's signal that the calls are independent.
Error handling is uniform across all language model providers:
'tool-not-available'andnotFoundistrue.error.