Initialize and configure Agent Control

Steps

Set the required environment variables

export GALILEO_API_KEY="<your-galileo-api-key>"

export GALILEO_CONSOLE_URL="https://console.<env>.galileo.ai"
export GALILEO_API_URL="https://api.<env>.galileo.ai"

export GALILEO_PROJECT="<project-name>"
export GALILEO_LOG_STREAM="<logstream-name>"

export AGENT_CONTROL_URL="https://agent-control.<env>.galileo.ai"
export AGENT_CONTROL_TARGET_TYPE="log_stream"
export AGENT_CONTROL_AGENT_NAME="default"
export AGENT_CONTROL_API_KEY_HEADER="Galileo-API-Key"

Initialize the Galileo Logger and the Agent Control SDK

Use the Galileo Logger around the agent run to write traces, spans, and control spans to the configured project and Log stream. The exact logger API may differ depending on your SDK version. You must initialize Galileo with the same project and Log stream that you plan to create controls for.The following example shows how to initialize the Galileo Logger.Note that:

The value for target_id should be logger.log_stream_id, not the Log stream name. The logger resolves the human-readable project and Log stream names to IDs, and Agent Control uses that resolved log_stream_id to fetch the controls bound in the console.
You must use the same API key unless your deployment has a separate Agent Control key. The key must be valid for your targeted Galileo environment.

import os
from uuid import uuid4

import agent_control
from galileo.logger.logger import GalileoLogger


def init_galileo_and_agent_control() -> GalileoLogger:
    # These names must match the project/Log stream where controls are bound.
    project_name = os.environ["GALILEO_PROJECT"]
    log_stream_name = os.environ["GALILEO_LOG_STREAM"]

    logger = GalileoLogger(
        project=project_name,
        log_stream=log_stream_name,
        mode=os.environ.get("GALILEO_LOGGER_MODE", "batch"),
    )

    logger.start_session(
        name=os.environ.get("GALILEO_SESSION_NAME", "agent-control-demo"),
        external_id=f"agent-control-demo-{uuid4()}",
        metadata={"source": "agent-control-demo"},
    )

    if logger.project_id is None or logger.log_stream_id is None:
        raise RuntimeError("Galileo logger did not resolve project/Log stream IDs.")

    # Useful for downstream SDK integrations/debugging.
    os.environ["GALILEO_PROJECT_ID"] = logger.project_id
    os.environ["GALILEO_LOG_STREAM_ID"] = logger.log_stream_id

    agent_control.init(
        agent_name=os.environ.get("AGENT_CONTROL_AGENT_NAME", "my-agent"),
        agent_description="Agent Control app with Galileo logging",
        server_url=os.environ["AGENT_CONTROL_URL"],
        api_key=os.environ["GALILEO_API_KEY"],
        api_key_header=os.environ.get("AGENT_CONTROL_API_KEY_HEADER", "Galileo-API-Key"),
        observability_enabled=True,
        observability_sink_name="registered",

        # Controls are usually bound to the Galileo Log stream.
        target_type=os.environ.get("AGENT_CONTROL_TARGET_TYPE", "log_stream"),
        target_id=logger.log_stream_id,
    )

    return logger

Run your agent with the Galileo Logger

logger = init_galileo_and_agent_control()

trace = logger.start_trace(
    name="my-agent-run",
    input={"user_request": "Wire $15,000 to Horizon Robotics."},
)

try:
    # Run your decorated agent Steps here.
    result = run_agent()
finally:
    logger.conclude(output={"result": result})
    logger.flush()

Decorate your LLM and tool calls

Wrap LLM and tool Steps with the SDK decorator so controls can evaluate the right Steps.

@control

Initialize and configure Agent Control

Overview

Before you begin

Steps

Next steps

​Overview

​Before you begin

​Steps

​Next steps

Overview

Before you begin

Steps

Next steps