tools.py
Tools allow your model to call external functions during training. Define them in tools.py using the @tool decorator.
Usage
from rnow.core.tool import tool
@tool
def search(query: str) -> dict:
"""Search for information."""
return {"results": ["result1", "result2"]}Requirements
docstring: Required, describes what the tool does
type hints: All parameters and return type must be typed
return: Must be JSON-serializable (str, int, float, bool, list, dict)
Supported Types
str, int, float, bool, list, dict
Examples
Calculator
from rnow.core.tool import tool
@tool
def calculator(expression: str) -> float:
"""Evaluate a math expression."""
return eval(expression)Web Search
import requests
from rnow.core.tool import tool
@tool
def web_search(query: str) -> dict:
"""Search the web."""
resp = requests.get("https://api.example.com/search", params={"q": query})
return resp.json()Sandbox Tools
Use @tool(sandbox=True) for tools that need isolated execution (code execution, file operations):
@tool(sandbox=True)
def run_python(code: str) -> str:
"""Execute Python code in isolated environment."""
exec(code)
return "Success"Entries using sandbox=True tools must have a docker field in train.jsonl specifying the container image.
Custom Docker images must be built for linux/amd64. Sandboxes run on x86_64 Linux servers:
docker build --platform linux/amd64 -t myorg/myimage:latest .Configuration
Enable multi-turn in config.yml:
rollout:
max_turns: 5
termination_policy: last_toollast_tool: Episode ends when model responds without a tool call
max_turns: Episode ends when max_turns is exhausted