gMAS / src /tools /function_calling.py
Артём Боярских
chore: initial commit
3193174
"""
Function calling tool — calling user-defined functions.
Allows agents to call registered Python functions.
Supports automatic schema extraction from type hints and docstrings.
"""
import inspect
from collections.abc import Callable
from typing import Any, Self, get_type_hints
from .base import BaseTool, ToolResult
def _python_type_to_json_schema(python_type: Any) -> dict[str, Any]:
"""Convert a Python type to a JSON Schema type."""
type_mapping = {
str: {"type": "string"},
int: {"type": "integer"},
float: {"type": "number"},
bool: {"type": "boolean"},
list: {"type": "array"},
dict: {"type": "object"},
type(None): {"type": "null"},
}
# Handle Optional, Union and other typing constructs
origin = getattr(python_type, "__origin__", None)
if origin is not None:
# List[X], dict[X, Y], etc.
if origin is list:
args = getattr(python_type, "__args__", ())
if args:
return {"type": "array", "items": _python_type_to_json_schema(args[0])}
return {"type": "array"}
if origin is dict:
return {"type": "object"}
# Union, Optional
return {"type": "string"} # fallback
return type_mapping.get(python_type, {"type": "string"})
def _extract_parameters_schema(func: Callable) -> dict[str, Any]:
"""Extract the JSON Schema of parameters from a function."""
try:
hints = get_type_hints(func)
except (NameError, AttributeError, TypeError):
hints = {}
sig = inspect.signature(func)
properties = {}
required = []
for param_name, param in sig.parameters.items():
if param_name in ("self", "cls"):
continue
param_type = hints.get(param_name, str)
if param_type is inspect.Parameter.empty:
param_type = str
prop = _python_type_to_json_schema(param_type)
# Try to extract description from docstring
# (simple heuristic: look for param_name: description)
properties[param_name] = prop
if param.default is inspect.Parameter.empty:
required.append(param_name)
schema = {
"type": "object",
"properties": properties,
}
if required:
schema["required"] = required
return schema
class FunctionWrapper(BaseTool):
"""
Wrapper for a Python function as a BaseTool.
Automatically extracts the name, description, and parameter schema from the function.
"""
def __init__(
self,
func: Callable,
tool_name: str | None = None,
tool_description: str | None = None,
):
"""
Create a wrapper for the function.
Args:
func: Python function.
tool_name: Tool name (defaults to the function name).
tool_description: Description (defaults to the docstring).
"""
self._func = func
self._name = tool_name or getattr(func, "__name__", "unnamed")
self._description = tool_description or getattr(func, "__doc__", None) or f"Function {self._name}"
self._parameters_schema = _extract_parameters_schema(func)
@property
def name(self) -> str:
return self._name
@property
def description(self) -> str:
return self._description
@property
def parameters_schema(self) -> dict[str, Any]:
return self._parameters_schema
def execute(self, **kwargs: Any) -> ToolResult:
"""Execute the function with the given arguments."""
try:
result = self._func(**kwargs)
output = str(result) if result is not None else "(no output)"
return ToolResult(
tool_name=self.name,
success=True,
output=output,
)
except (TypeError, ValueError, KeyError, AttributeError) as e:
return ToolResult(
tool_name=self.name,
success=False,
error=f"Execution error: {e}",
)
class FunctionTool(BaseTool):
"""
Manager of user-defined functions as a tool.
Allows registering multiple functions and calling them by name.
This is a meta-tool that manages a collection of functions.
Example:
tool = FunctionTool()
@tool.register
def calculate(expression: str) -> str:
\"\"\"Evaluate a mathematical expression.\"\"\"
return str(eval(expression))
@tool.register
def uppercase(text: str) -> str:
\"\"\"Convert text to upper case.\"\"\"
return text.upper()
# Call a specific function
result = tool.execute(function="calculate", expression="2 + 2")
"""
def __init__(self):
"""Create an empty FunctionTool."""
self._functions: dict[str, Callable] = {}
self._wrappers: dict[str, FunctionWrapper] = {}
@property
def name(self) -> str:
return "function_calling"
@property
def description(self) -> str:
func_list = ", ".join(self._functions.keys()) if self._functions else "none"
return f"Call a registered function. Available functions: {func_list}"
@property
def parameters_schema(self) -> dict[str, Any]:
return {
"type": "object",
"properties": {
"function": {
"type": "string",
"description": "Name of the function to call",
"enum": list(self._functions.keys()) if self._functions else [],
},
},
"required": ["function"],
"additionalProperties": True, # Allow function-specific args
}
def register(
self,
func: Callable | None = None,
*,
name: str | None = None,
description: str | None = None,
) -> Callable:
"""
Register a function. Can be used as a decorator.
Args:
func: Function to register.
name: Function name (defaults to func.__name__).
description: Description (defaults to the docstring).
Returns:
The original function.
Example:
@tool.register
def my_function(arg: str) -> str:
return arg
@tool.register(name="custom_name")
def another(x: int) -> int:
return x * 2
"""
def decorator(f: Callable) -> Callable:
func_name = name or getattr(f, "__name__", "unnamed")
self._functions[func_name] = f
self._wrappers[func_name] = FunctionWrapper(
func=f,
tool_name=func_name,
tool_description=description or f.__doc__,
)
return f
if func is not None:
return decorator(func)
return decorator
def add_function(
self,
func: Callable,
name: str | None = None,
description: str | None = None,
) -> Self:
"""
Add a function (non-decorator way).
Args:
func: Function to add.
name: Function name.
description: Function description.
Returns:
self for method chaining.
"""
self.register(func, name=name, description=description)
return self
def get_function(self, name: str) -> Callable | None:
"""Get a function by name."""
return self._functions.get(name)
def list_functions(self) -> list[str]:
"""Get the list of registered function names."""
return list(self._functions.keys())
def execute(self, function: str = "", **kwargs: Any) -> ToolResult:
"""
Execute a registered function.
Args:
function: Name of the function to call.
**kwargs: Function arguments.
Returns:
ToolResult with the result.
"""
if not function:
return ToolResult(
tool_name=self.name,
success=False,
error="No function name provided",
)
wrapper = self._wrappers.get(function)
if wrapper is None:
available = ", ".join(self._functions.keys()) or "none"
return ToolResult(
tool_name=self.name,
success=False,
error=f"Function '{function}' not found. Available: {available}",
)
# Call via wrapper (which is already a BaseTool)
return wrapper.execute(**kwargs)