Agent Commands
There is no buddy agent group
The buddy CLI has no dedicated command group for running agents. You serve
agents from Python (via the app classes) or, for declarative
deployments, through the workspace resources system. The CLI's
top-level commands and buddy ws handle the
latter.
Serving an agent from Python
The most direct path is to build an app object and call serve(). For a REST
API:
from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.app.fastapi import FastAPIApp
agent = Agent(agent_id="api_agent", model=OpenAIChat(id="gpt-4o-mini"))
fastapi_app = FastAPIApp(agents=[agent])
app = fastapi_app.get_app()
if __name__ == "__main__":
fastapi_app.serve(app, host="0.0.0.0", port=7777)
Run it like any script:
See FastAPI Apps for the endpoints and request shape.
Interactive playground
To chat with and debug an agent in the browser:
from buddy.app.playground import Playground
playground = Playground(agents=[agent])
app = playground.get_app()
playground.serve(app, host="localhost", port=7777)
A Playground URL pointing at your local endpoint is printed to the console.
Declarative deployment via a workspace
If you manage infrastructure with a resources.py file, start and stop your
agent resources through the CLI:
buddy start # start resources defined in resources.py
buddy stop # stop them
# or, for the active workspace:
buddy ws up
buddy ws down
See Workspace Management for filters and options.
Related CLI groups
buddy train— train a local model, then back an agent with it usinguse_with_agent(see Model Training)buddy pulse— run the PULSE virtual employee