Quick start

As a library

from pycodeloop import CodeLoop, Config
from pycodeloop.providers import GenericProvider

config = Config(provider=GenericProvider.from_json("templates/anthropic.json"))
flow = CodeLoop(config=config)

print(flow.run("list the files in this repo and summarize the project"))

From the CLI

pycodeloop run "add a docstring to pycodeloop/core/agent.py"  # one-shot, scriptable
pycodeloop                                                 # full-screen chat (the default with no args)

Low-level: Agent directly

CodeLoop is a convenience wrapper. Reach for Agent directly when you want per-call control over hooks, tools, or session state:

from pycodeloop.core.agent import Agent
from pycodeloop.providers import GenericProvider

agent = Agent(
    provider=GenericProvider.from_json("templates/anthropic.json"),
    on_tool_call=lambda name, args: print(f"-> {name} {args}"),
)

reply = agent.run("fix the failing test in tests/test_agent.py")

Next: Streaming, Permission prompts, Token usage.