Skip to main content
YAML tools: under framework: autogen (v0.2) are now fully wired into the AutoGen chat — the assistant advertises them to the LLM and the user_proxy executes them.

​
Quick Start

1

Declare the tool in YAML

Reference the tool by name under a role’s tools: list.
framework: autogen
topic: Weather report

roles:
  weather_agent:
    role: Meteorologist
    goal: Report weather for a city
    backstory: Reliability-focused agent.
    tools:
      - get_weather
    tasks:
      report:
        description: Report the weather for {topic}
        expected_output: One paragraph forecast.
2

Define the tool and run

Define get_weather as a plain callable with a __name__ and a docstring, then run the YAML.
from praisonai import PraisonAI

def get_weather(city: str) -> str:
    """Return the current weather for a city."""
    return f"It is sunny and 22°C in {city}."

praison = PraisonAI(agent_file="agents.yaml", tools=[get_weather])
praison.run()

​
How It Works

The adapter registers each callable on both AutoGen v0.2 agents so the LLM can call it and the user_proxy can run it.
StepMethodPurpose
Advertise schemaassistant.register_for_llm(name, description)The LLM learns the tool exists
Execute calluser_proxy.register_for_execution(name)The user_proxy runs the callable
DescriptionFirst line of __doc__, else Tool <name>Shown to the LLM

​
Skip Rules

Tools that cannot be wired are logged and skipped — never silently dropped.
SituationBehaviour
Non-callable toolSkipped with a warning
No resolvable __name__ / nameSkipped with a warning
Duplicate tool name on the shared user_proxyFirst callable kept, duplicate skipped with a warning
Duplicate handling prevents AutoGen’s last-write-wins _function_map from silently overwriting an earlier tool when two agents declare a same-named callable.

​
Timeout Translation

AutoGen v0.2 runs tools inside its chat loop and expects a string to hand back to the LLM. When a per-call tool timeout raises ToolTimeoutError, the adapter’s _wrap_tool_for_execution catches it and returns "Error: tool '<name>' timed out (…)." so the conversation keeps moving instead of aborting. See Tool Configuration → Wrapper-Level Tool Timeout for the full timeout contract.

​
Best Practices

The tool name comes from __name__ (or a name attribute), and the LLM-facing description is the first line of the docstring. Named functions with a one-line docstring wire cleanly.
def get_weather(city: str) -> str:
    """Return the current weather for a city."""
    return f"Sunny in {city}."
Lambdas have no __name__, so the adapter cannot resolve a name and skips them with a warning. Use def.
Two agents declaring a tool with the same __name__ share one user_proxy; only the first callable is registered for execution. Give each tool a distinct name.

AutoGen Tool Registry Migration

Migrate tool registries into AutoGen

AutoGen LLM config_list

Configure the AutoGen LLM config_list