Model Response Overview

Output control makes results structured and reliable. This chapter explains how to read and consume those results. Agently wraps model returns into a unified response object so you can access text, structured data, metadata, and streaming events consistently.

Response pipeline

Key objects

• ModelResponse: a response snapshot with response.id, plus prompt and settings snapshots. Use cancel_logs() to stop logs.
• ModelResponseResult: result access via get_text() / get_data() / get_data_object() / get_meta() / get_generator().
• ResponseParser: normalizes the event stream and parses structured output. Default: AgentlyResponseParser.

Recommended usage: keep one response

In chained calls, agent.start() / agent.get_text() / agent.get_data() / agent.get_data_object() / agent.get_meta() / agent.get_generator() (and the same methods on a request) implicitly create a new response and trigger a new request. To keep a single request fixed, call get_response() first, then read text, structured data, metadata, or streaming events from the response as many times as needed.

from agently import Agently

agent = Agently.create_agent()

response = (
  agent
  .input("Introduce Agently in one sentence")
  .output({
    "intro": ("str", "One-line intro"),
  })
  .get_response()
)

text = response.get_text()
data = response.get_data()
meta = response.get_meta()

print(text)
print(data)
print(meta)