Skip to main content
from praisonaiagents import Agent

agent = Agent(name="assistant", instructions="Be helpful.", output=True)
agent.start("Explain loop detection in one paragraph.")
The display system formats agent output in the terminal and exposes hooks for custom rendering.
This page covers terminal output only. For per-channel presentation (streaming, tool progress, footers), see Display Policy.
The user runs the agent in a terminal; the display system renders Rich output and optional global callbacks.

​
How It Works

​
Quick Start

1

Default terminal output

from praisonaiagents import Agent

agent = Agent(
    name="assistant",
    instructions="Be helpful.",
    output=True,
)
agent.start("Explain display callbacks in one sentence.")
2

Custom interaction handler

from praisonaiagents import Agent, register_display_callback

def compact(message=None, response=None, **kwargs):
    name = kwargs.get("agent_name", "agent")
    print(f"[{name}] {response}")

register_display_callback("interaction", compact)

agent = Agent(name="writer", instructions="Keep answers short.")
agent.start("One tip for readable logs.")
3

Inspect TaskOutput in a task callback

from praisonaiagents import Agent, Task, TaskOutput, PraisonAIAgents

def show_result(output: TaskOutput):
    print(output.raw)
    if output.metadata:
        print("meta:", output.metadata)

agent = Agent(name="analyst", instructions="Analyse briefly.")
task = Task(
    description="List three benefits of typed task output",
    expected_output="Bulleted list",
    agent=agent,
    callback=show_result,
)
PraisonAIAgents(agents=[agent], tasks=[task]).start()

​
Core Types

​
TaskOutput

FieldTypeDescription
rawstrText result from the task
json_outputdictParsed JSON when structured output is enabled
pydantic_outputBaseModelParsed model when a schema is set
task_idstrTask identifier
metadatadictExecution metrics and context

​
ReflectionOutput

FieldTypeDescription
reflectionstrSelf-reflection content
satisfactoryboolWhether reflection passed
improvement_suggestionslistSuggested improvements
Import both from the top level: from praisonaiagents import TaskOutput, ReflectionOutput.

​
Global Registries

RegistryPurpose
sync_display_callbacksSync handlers keyed by event type
async_display_callbacksAsync handlers keyed by event type
register_display_callbackRegister into either registry
Built-in display helpers (display_interaction, display_tool_call, display_error, and others) call registered callbacks even when verbose output is off.

​
Best Practices

Register handlers once at startup instead of monkey-patching internal display functions.
When you need the full TaskOutput object, attach a callback on Task rather than an interaction handler.
Set output=False in CI or server mode and rely on callbacks for structured logs.
Event names, autonomy callbacks, and async patterns are documented on the Display Callbacks page.

Display Callbacks

Event types and registration patterns

Output Config

Verbose, stream, and markdown output settings