Tools

Tools are how the model performs actions in the workspace.

Built-in core tools

These are enabled by default:

• bash: run a shell command.
• params: cmd: string
• write_file: create/overwrite files in workspace.
• params: path: string, content: string
• read_image: read an image file and return a data URL.
• params: path: string
• ask_user: request additional user input.
• stop: signal the autonomous loop to stop.

Optional legacy task tools

If you want delegated multi-agent flows, register them explicitly:

from pygent.task_tools import register_task_tools

register_task_tools()

Additional tools then become available:

• delegate_task
• delegate_persona_task
• task_status
• collect_file
• list_personas
• download_file

Registering custom tools

from pygent import Agent, register_tool

def hello(rt, name: str) -> str:
    return f"Hello {name}!"

register_tool(
    "hello",
    "Greet by name",
    {"type": "object", "properties": {"name": {"type": "string"}}, "required": ["name"]},
    hello,
)

ag = Agent()
ag.step("hello name='world'")

Decorator version:

from pygent import Agent, tool

@tool(
    name="goodbye",
    description="Say goodbye",
    parameters={"type": "object", "properties": {"name": {"type": "string"}}, "required": ["name"]},
)
def goodbye(rt, name: str) -> str:
    return f"Goodbye {name}!"