Running Agents

Learn how to run Agno Agents.

Run your Agent by calling Agent.run() or Agent.arun(). Here’s how they work:

1. The agent builds the context to send to the model (system message, user message, chat history, user memories, session state and other relevant inputs).
2. The agent sends this context to the model.
3. The model processes the input and responds with either a message or a tool call.
4. If the model makes a tool call, the agent executes it and returns the results to the model.
5. The model processes the updated context, repeating this loop until it produces a final message without any tool calls.
6. The agent returns this final response to the caller.

Basic Execution

The Agent.run() function runs the agent and returns the output — either as a RunOutput object or as a stream of RunOutputEvent objects (when stream=True). For example:

from agno.agent import Agent, RunOutput
from agno.models.anthropic import Claude
from agno.tools.hackernews import HackerNewsTools
from agno.utils.pprint import pprint_run_response

agent = Agent(
    model=Claude(id="claude-sonnet-4-5"),
    tools=[HackerNewsTools()],
    instructions="Write a report on the topic. Output only the report.",
    markdown=True,
)

# Run agent and return the response as a variable
response: RunOutput = agent.run("Trending startups and products.")
# Print the response in markdown format
pprint_run_response(response, markdown=True)

You can also run the agent asynchronously using Agent.arun(). See this example.

Run Input

The input parameter is the input to send to the agent. It can be a string, a list, a dictionary, a message, a pydantic model or a list of messages. For example:

from agno.agent import Agent, RunOutput
from agno.models.anthropic import Claude
from agno.tools.hackernews import HackerNewsTools
from agno.utils.pprint import pprint_run_response

agent = Agent(
    model=Claude(id="claude-sonnet-4-5"),
    tools=[HackerNewsTools()],
    instructions="Write a report on the topic. Output only the report.",
    markdown=True,
)

# Run agent with input="Trending startups and products."
response: RunOutput = agent.run(input="Trending startups and products.")
# Print the response in markdown format
pprint_run_response(response, markdown=True)

See the Input & Output docs for more information, and to see how to use structured input and output with agents.

Run Output

The Agent.run() function returns a RunOutput object when not streaming. Here are some of the core attributes:

• run_id: The id of the run.
• agent_id: The id of the agent.
• agent_name: The name of the agent.
• session_id: The id of the session.
• user_id: The id of the user.
• content: The response content.
• content_type: The type of content. In the case of structured output, this will be the class name of the pydantic model.
• reasoning_content: The reasoning content.
• messages: The list of messages sent to the model.
• metrics: The metrics of the run. For more details see Metrics.
• model: The model used for the run.

See detailed documentation in the RunOutput documentation.

Streaming

To enable streaming, set stream=True when calling run(). This will return an iterator of RunOutputEvent objects instead of a single response.

from typing import Iterator
from agno.agent import Agent, RunOutputEvent, RunEvent
from agno.models.anthropic import Claude
from agno.tools.hackernews import HackerNewsTools

agent = Agent(
    model=Claude(id="claude-sonnet-4-5"),
    tools=[HackerNewsTools()],
    instructions="Write a report on the topic. Output only the report.",
    markdown=True,
)

# Run agent and return the response as a stream
stream: Iterator[RunOutputEvent] = agent.run("Trending products", stream=True)
for chunk in stream:
    if chunk.event == RunEvent.run_content:
        print(chunk.content)

For asynchronous streaming, see this example.

Streaming all events

By default, when you stream a response, only the RunContent events will be streamed.

You can also stream all run events by setting stream_events=True.

This will provide real-time updates about the agent’s internal processes, like tool calling or reasoning:

. For example:

# Stream all events
response_stream: Iterator[RunOutputEvent] = agent.run(
    "Trending products",
    stream=True,
    stream_events=True
)

Handling Events

You can process events as they arrive by iterating over the response stream:

from agno.agent import Agent, RunEvent
from agno.models.anthropic import Claude
from agno.tools.hackernews import HackerNewsTools

agent = Agent(
    model=Claude(id="claude-sonnet-4-5"),
    tools=[HackerNewsTools()],
    instructions="Write a report on the topic. Output only the report.",
    markdown=True,
)

stream = agent.run("Trending products", stream=True, stream_events=True)

for chunk in stream:
    if chunk.event == RunEvent.run_content:
        print(f"Content: {chunk.content}")
    elif chunk.event == RunEvent.tool_call_started:
        print(f"Tool call started: {chunk.tool.tool_name}")
    elif chunk.event == RunEvent.reasoning_step:
        print(f"Reasoning step: {chunk.content}")

RunEvents make it possible to build exceptional agent experiences.

Event Types

The following events are yielded by the Agent.run() and Agent.arun() functions depending on the agent’s configuration:

Core Events

Event Type	Description
RunStarted	Indicates the start of a run
RunContent	Contains the model’s response text as individual chunks
RunContentCompleted	Signals completion of content streaming
RunIntermediateContent	Contains the model’s intermediate response text as individual chunks. This is used when output_model is set.
RunCompleted	Signals successful completion of the run
RunError	Indicates an error occurred during the run
RunCancelled	Signals that the run was cancelled

Control Flow Events

Event Type	Description
RunPaused	Indicates the run has been paused
RunContinued	Signals that a paused run has been continued

Tool Events

Event Type	Description
ToolCallStarted	Indicates the start of a tool call
ToolCallCompleted	Signals completion of a tool call, including tool call results

Reasoning Events

Event Type	Description
ReasoningStarted	Indicates the start of the agent’s reasoning process
ReasoningStep	Contains a single step in the reasoning process
ReasoningCompleted	Signals completion of the reasoning process

Memory Events

Event Type	Description
MemoryUpdateStarted	Indicates that the agent is updating its memory
MemoryUpdateCompleted	Signals completion of a memory update

Session Summary Events

Event Type	Description
SessionSummaryStarted	Indicates the start of session summary generation
SessionSummaryCompleted	Signals completion of session summary generation

Pre-Hook Events

Event Type	Description
PreHookStarted	Indicates the start of a pre-run hook
PreHookCompleted	Signals completion of a pre-run hook execution

Post-Hook Events

Event Type	Description
PostHookStarted	Indicates the start of a post-run hook
PostHookCompleted	Signals completion of a post-run hook execution

Parser Model events

Event Type	Description
ParserModelResponseStarted	Indicates the start of the parser model response
ParserModelResponseCompleted	Signals completion of the parser model response

Output Model events

Event Type	Description
OutputModelResponseStarted	Indicates the start of the output model response
OutputModelResponseCompleted	Signals completion of the output model response

Custom Events

If you are using your own custom tools, you can yield custom events along with the rest of the Agno events. Create a custom event class by extending the CustomEvent class. For example:

from dataclasses import dataclass
from agno.run.agent import CustomEvent

@dataclass
class CustomerProfileEvent(CustomEvent):
    """CustomEvent for customer profile."""

    customer_name: Optional[str] = None
    customer_email: Optional[str] = None
    customer_phone: Optional[str] = None

You can then yield your custom event from your tool. The event will be handled internally as an Agno event, and you will be able to access it in the same way you would access any other Agno event. For example:

from agno.tools import tool

@tool()
async def get_customer_profile():
    """Example custom tool that simply yields a custom event."""

    yield CustomerProfileEvent(
        customer_name="John Doe",
        customer_email="[e-mail]",
        customer_phone="1234567890",
    )

See the full example for more details.

Specify Run User and Session

You can specify which user and session to use when running the agent by passing the user_id and session_id parameters. This ensures the current run is associated with the correct user and session. For example:

agent.run("Tell me a 5 second short story about a robot", user_id="[e-mail]", session_id="session_123")

For more information see the Agent Sessions documentation.

Passing Images / Audio / Video / Files

You can pass images, audio, video, or files to the agent by passing the images, audio, video, or files parameters. For example:

agent.run("Tell me a 5 second short story about this image", images=[Image(url="https://example.com/image.jpg")])

For more information see the Multimodal Agents documentation.

Pausing and Continuing a Run

An agent run can be paused when a human-in-the-loop flow is initiated. You can then continue the execution of the agent by calling the Agent.continue_run() method.

See more details in the Human-in-the-Loop documentation.

Cancelling a Run

A run can be cancelled by calling the Agent.cancel_run() method.

See more details in the Cancelling a Run documentation.

Developer Resources

• View the Agent reference
• View the RunOutput schema
• View Agent Cookbook