set_gather_info | SignalWire

Enable info gathering for this step. Call add_gather_question() after this method to define the questions.

The gather_info system collects structured information from the caller by presenting questions one at a time. It uses dynamic step instruction re-injection rather than tool calls, producing zero tool_call/tool_result entries in LLM-visible history.

Parameters

output_key

str | NoneDefaults to None

Key in global_data to store collected answers under. When None, answers are stored at the top level of global_data.

completion_action

str | NoneDefaults to None

Where to go when all questions are answered.

"next_step" — auto-advance to the next sequential step
A step name (e.g., "process_results") — jump to that specific step
None — return to normal step mode after gathering

prompt

str | NoneDefaults to None

Preamble text injected once when entering the gather step, giving the AI personality and context for asking the questions.

isolated

boolDefaults to False

Default for every question in this gather. When True, each question is asked with the sibling questions and answers hidden from the model, so it must ask rather than derive the answer from an earlier one. A question’s own isolated overrides this. Hidden turns stay in the call log.

Returns

Step — Self for method chaining.

Example

from signalwire import AgentBase
agent = AgentBase(name="my-agent", route="/agent")
contexts = agent.define_contexts()
ctx = contexts.add_context("default")
intake = ctx.add_step("intake")
intake.set_text("Collect patient information.")
intake.set_gather_info(
    output_key="patient_info",
    completion_action="next_step",
    prompt="Be friendly and professional when collecting information."
)
intake.add_gather_question(
    key="full_name",
    question="What is your full name?",
    confirm=True
)
ctx.add_step("review").set_text("Review the collected information with the patient.")
agent.serve()