Tool Decorator

🔧 You are viewing: API Specification - Complete technical reference for the @tool decoratorSee also: Conceptual Guide

Component: Tool Decorator
Module: gaia.agents.base.tools
Import: from gaia.agents.base.tools import tool
Source: src/gaia/agents/base/tools.py

Overview

The @tool decorator is a simple, powerful mechanism for registering Python functions as agent tools. It automatically extracts function signatures, type annotations, and docstrings to create tool definitions that LLMs can understand and use. What it does:

Registers functions in the global tool registry
Extracts function signature and parameters
Infers parameter types from type annotations
Captures docstrings for tool descriptions
Generates JSON schema for LLM function calling
Supports both @tool and @tool() syntax

Why use it:

Minimal boilerplate (single decorator)
Automatic type inference from Python annotations
Clear separation between tool definition and implementation
Compatible with standard Python typing

Purpose and Use Cases

When to Use

Registering Agent Tools
- Every function you want the agent to call
- Tools that perform actions (database queries, file operations, API calls)
- Tools that retrieve information (search, list, get status)
Simple Tool Definition
- Quick tool prototyping
- Small to medium agents with 3-20 tools
- Standard Python function signatures
Type-Safe Tools
- Tools with clear parameter types
- Tools requiring type validation
- Tools used by multiple agents

When NOT to Use

Functions not meant to be called by the agent
Internal helper methods
Methods that require agent instance state (use instance methods in _register_tools() instead)

API Specification

Decorator Definition

from typing import Callable

def tool(
    func: Callable = None,
    **kwargs  # Ignored, for backward compatibility
) -> Callable:
    """
    Decorator to register a function as an agent tool.

    Supports both @tool and @tool(...) syntax.
    Automatically extracts:
    - Function name
    - Parameter names, types, and defaults
    - Docstring for description

    Args:
        func: Function to register (when used as @tool)
        **kwargs: Optional arguments (ignored)

    Returns:
        Original function, unchanged

    Example:
        @tool
        def search(query: str, limit: int = 10) -> dict:
            '''Search for items.

            Args:
                query: Search query string
                limit: Max results to return

            Returns:
                Search results
            '''
            return {"results": [...]}
    """

Tool Registry

# Global tool registry (internal use)
_TOOL_REGISTRY: Dict[str, Dict[str, Any]] = {}

# Registry structure:
{
    "tool_name": {
        "name": str,              # Function name
        "description": str,       # From docstring
        "parameters": {
            "param_name": {
                "type": str,      # "string", "integer", "number", "boolean", "array", "object", "unknown"
                "required": bool, # True if no default value
            }
        },
        "function": Callable      # Original function
    }
}

Type Mapping

# Python type → JSON schema type
str → "string"
int → "integer"
float → "number"
bool → "boolean"
tuple → "array"
dict → "object"
Dict → "object"
(no annotation) → "unknown"

Code Examples

Example 1: Basic Tool

from gaia.agents.base.tools import tool

@tool
def calculate(expression: str) -> float:
    """
    Calculate a mathematical expression.

    Args:
        expression: Math expression to evaluate (e.g., "2 + 2")

    Returns:
        Calculated result
    """
    import math
    # Safe evaluation in production
    return eval(expression, {"__builtins__": {}}, math.__dict__)

# Registry entry created:
{
    "calculate": {
        "name": "calculate",
        "description": "Calculate a mathematical expression...",
        "parameters": {
            "expression": {
                "type": "string",
                "required": True
            }
        },
        "function": <function calculate>
    }
}

Example 2: Tool with Optional Parameters

@tool
def search_users(
    query: str,
    limit: int = 10,
    include_inactive: bool = False
) -> dict:
    """
    Search for users in the database.

    Args:
        query: Search query string
        limit: Maximum number of results (default: 10)
        include_inactive: Include inactive users (default: False)

    Returns:
        Dictionary with users and count
    """
    # Implementation
    return {"users": [], "count": 0}

# Registry entry:
{
    "search_users": {
        "name": "search_users",
        "description": "Search for users in the database...",
        "parameters": {
            "query": {"type": "string", "required": True},
            "limit": {"type": "integer", "required": False},
            "include_inactive": {"type": "boolean", "required": False}
        },
        "function": <function search_users>
    }
}

Example 3: Tool with Complex Types

from typing import Dict, List

@tool
def batch_process(
    items: tuple,
    config: dict,
    dry_run: bool = False
) -> dict:
    """
    Process multiple items with configuration.

    Args:
        items: List of items to process
        config: Configuration dictionary
        dry_run: Run without making changes

    Returns:
        Processing results
    """
    return {"processed": len(items), "dry_run": dry_run}

# Registry entry:
{
    "batch_process": {
        "name": "batch_process",
        "description": "Process multiple items...",
        "parameters": {
            "items": {"type": "array", "required": True},
            "config": {"type": "object", "required": True},
            "dry_run": {"type": "boolean", "required": False}
        },
        "function": <function batch_process>
    }
}

Example 4: Tool in Agent Context

from gaia.agents.base.agent import Agent
from gaia.agents.base.tools import tool

class MyAgent(Agent):
    def __init__(self, db_path: str, **kwargs):
        self.db_path = db_path
        super().__init__(**kwargs)

    def _register_tools(self):
        # Tool with access to instance state
        @tool
        def query_db(sql: str) -> dict:
            """Execute SQL query on database.

            Args:
                sql: SQL query string

            Returns:
                Query results
            """
            # Can access self.db_path here
            import sqlite3
            conn = sqlite3.connect(self.db_path)
            cursor = conn.execute(sql)
            results = [dict(row) for row in cursor.fetchall()]
            conn.close()
            return {"results": results, "count": len(results)}

        @tool
        def list_tables() -> dict:
            """List all tables in the database.

            Returns:
                List of table names
            """
            import sqlite3
            conn = sqlite3.connect(self.db_path)
            cursor = conn.execute(
                "SELECT name FROM sqlite_master WHERE type='table'"
            )
            tables = [row[0] for row in cursor.fetchall()]
            conn.close()
            return {"tables": tables}

Example 5: Both Decorator Syntaxes

# Syntax 1: No parentheses (recommended)
@tool
def simple_tool(param: str) -> dict:
    """Simple tool."""
    return {"result": param}

# Syntax 2: With parentheses (backward compatibility)
@tool()
def another_tool(param: str) -> dict:
    """Another tool."""
    return {"result": param}

# Both work identically

Implementation Details

Type Inference

def decorator(f: Callable) -> Callable:
    sig = inspect.signature(f)
    params = {}

    for name, param in sig.parameters.items():
        param_info = {
            "type": "unknown",
            "required": param.default == inspect.Parameter.empty
        }

        # Infer type from annotation
        if param.annotation != inspect.Parameter.empty:
            if param.annotation == str:
                param_info["type"] = "string"
            elif param.annotation == int:
                param_info["type"] = "integer"
            elif param.annotation == float:
                param_info["type"] = "number"
            elif param.annotation == bool:
                param_info["type"] = "boolean"
            elif param.annotation == tuple:
                param_info["type"] = "array"
            elif param.annotation == dict or param.annotation == Dict:
                param_info["type"] = "object"

        params[name] = param_info

    return params

Docstring Parsing

# Extracts entire docstring as-is
description = f.__doc__ or ""

# No special parsing of Args/Returns sections
# LLM reads the full docstring

Registry Management

# Global registry (cleared when agent reinitializes)
_TOOL_REGISTRY = {}

# Tools registered at decoration time
_TOOL_REGISTRY[tool_name] = {
    "name": tool_name,
    "description": f.__doc__ or "",
    "parameters": params,
    "function": f
}

Testing Requirements

Unit Tests

File: tests/agents/base/test_tools.py

import pytest
from gaia.agents.base.tools import tool, _TOOL_REGISTRY

def test_tool_decorator_basic():
    """Test basic tool registration."""
    # Clear registry
    _TOOL_REGISTRY.clear()

    @tool
    def test_func(param: str) -> dict:
        """Test function."""
        return {"result": param}

    assert "test_func" in _TOOL_REGISTRY
    assert _TOOL_REGISTRY["test_func"]["name"] == "test_func"
    assert _TOOL_REGISTRY["test_func"]["description"] == "Test function."

def test_tool_parameter_types():
    """Test type inference."""
    _TOOL_REGISTRY.clear()

    @tool
    def typed_func(
        s: str,
        i: int,
        f: float,
        b: bool,
        t: tuple,
        d: dict
    ) -> dict:
        """Function with types."""
        return {}

    params = _TOOL_REGISTRY["typed_func"]["parameters"]
    assert params["s"]["type"] == "string"
    assert params["i"]["type"] == "integer"
    assert params["f"]["type"] == "number"
    assert params["b"]["type"] == "boolean"
    assert params["t"]["type"] == "array"
    assert params["d"]["type"] == "object"

def test_tool_optional_parameters():
    """Test required vs optional parameters."""
    _TOOL_REGISTRY.clear()

    @tool
    def optional_func(required: str, optional: int = 10) -> dict:
        """Function with optional param."""
        return {}

    params = _TOOL_REGISTRY["optional_func"]["parameters"]
    assert params["required"]["required"] is True
    assert params["optional"]["required"] is False

def test_tool_with_parentheses():
    """Test @tool() syntax."""
    _TOOL_REGISTRY.clear()

    @tool()
    def paren_func(param: str) -> dict:
        """Parentheses syntax."""
        return {}

    assert "paren_func" in _TOOL_REGISTRY

def test_tool_function_unchanged():
    """Test decorated function still works."""
    @tool
    def add(a: int, b: int) -> int:
        """Add two numbers."""
        return a + b

    # Function still callable
    result = add(2, 3)
    assert result == 5

def test_tool_no_annotations():
    """Test tool without type annotations."""
    _TOOL_REGISTRY.clear()

    @tool
    def no_types(param):
        """No type annotations."""
        return {"result": param}

    params = _TOOL_REGISTRY["no_types"]["parameters"]
    assert params["param"]["type"] == "unknown"
    assert params["param"]["required"] is True

def test_tool_no_docstring():
    """Test tool without docstring."""
    _TOOL_REGISTRY.clear()

    @tool
    def no_doc(param: str) -> dict:
        return {"result": param}

    assert _TOOL_REGISTRY["no_doc"]["description"] == ""

Dependencies

Required Packages

import inspect
import logging
from typing import Callable, Dict

No External Dependencies

The tool decorator uses only Python standard library.

Acceptance Criteria

Best Practices

Good Tool Docstrings

@tool
def process_data(data_id: str, operation: str = "validate") -> dict:
    """
    Process a data item with specified operation.

    Performs validation, transformation, or analysis on data items.
    Default operation is validation.

    Args:
        data_id: Unique identifier of the data item
        operation: Type of processing ("validate", "transform", "analyze")

    Returns:
        dict: {
            "status": "success" or "error",
            "data_id": str,
            "result": Any,
            "error": str (if status is "error")
        }

    Example:
        result = process_data("item-123", "validate")
        # Returns: {"status": "success", "data_id": "item-123", "result": {...}}
    """

Clear Parameter Names

# Good: Clear intent
@tool
def search_customer(name_query: str, max_results: int = 10) -> dict:
    """Search customers by name."""

# Bad: Unclear names
@tool
def search_customer(q: str, n: int = 10) -> dict:
    """Search customers."""

Consistent Return Types

# Good: Always return dict
@tool
def get_user(user_id: int) -> dict:
    """Get user by ID."""
    return {"user": {...}, "found": True}

# Good: Consistent structure
@tool
def create_user(name: str) -> dict:
    """Create new user."""
    return {"user_id": 123, "status": "created"}

Limitations

No Advanced Type Hints

# Not supported: Union types
def process(data: Union[str, int]) -> dict:  # type becomes "unknown"

# Not supported: Optional
def process(data: Optional[str]) -> dict:  # type becomes "unknown"

# Not supported: List[str]
def process(items: List[str]) -> dict:  # type is "array" but not List[str]

# Workaround: Use simpler types
def process(data: str) -> dict:  # Documented in docstring

No Parameter Validation

# Decorator doesn't validate arguments
@tool
def divide(a: int, b: int) -> float:
    """Divide a by b."""
    return a / b  # No check for b == 0

# Add validation in function body
@tool
def divide(a: int, b: int) -> dict:
    """Divide a by b."""
    if b == 0:
        return {"error": "Division by zero"}
    return {"result": a / b}

Future Enhancements

Support for Union types
Support for Optional types
Support for List[Type], Dict[Key, Value]
Pydantic model integration
Parameter validation decorators
Tool categories/tags
Tool versioning
Automatic example generation from docstring

@tool Decorator Technical Specification

Core Framework

SDKs

Infrastructure

Code Infrastructure

Tool Mixins

Packaging

Agents & Apps

Overview

Purpose and Use Cases

When to Use

When NOT to Use

API Specification

Decorator Definition

Tool Registry

Type Mapping

Code Examples

Example 1: Basic Tool

Example 2: Tool with Optional Parameters

Example 3: Tool with Complex Types

Example 4: Tool in Agent Context

Example 5: Both Decorator Syntaxes

Implementation Details

Type Inference

Docstring Parsing

Registry Management

Testing Requirements

Unit Tests

Dependencies

Required Packages

No External Dependencies

Acceptance Criteria

Best Practices

Good Tool Docstrings

Clear Parameter Names

Consistent Return Types

Limitations

No Advanced Type Hints

No Parameter Validation

Future Enhancements

Core Framework

SDKs

Infrastructure

Code Infrastructure

Tool Mixins

Packaging

Agents & Apps

​Overview

​Purpose and Use Cases

​When to Use

​When NOT to Use

​API Specification

​Decorator Definition

​Tool Registry

​Type Mapping

​Code Examples

​Example 1: Basic Tool

​Example 2: Tool with Optional Parameters

​Example 3: Tool with Complex Types

​Example 4: Tool in Agent Context

​Example 5: Both Decorator Syntaxes

​Implementation Details

​Type Inference

​Docstring Parsing

​Registry Management

​Testing Requirements

​Unit Tests

​Dependencies

​Required Packages

​No External Dependencies

​Acceptance Criteria

​Best Practices

​Good Tool Docstrings

​Clear Parameter Names

​Consistent Return Types

​Limitations

​No Advanced Type Hints

​No Parameter Validation

​Future Enhancements

Overview

Purpose and Use Cases

When to Use

When NOT to Use

API Specification

Decorator Definition

Tool Registry

Type Mapping

Code Examples

Example 1: Basic Tool

Example 2: Tool with Optional Parameters

Example 3: Tool with Complex Types

Example 4: Tool in Agent Context

Example 5: Both Decorator Syntaxes

Implementation Details

Type Inference

Docstring Parsing

Registry Management

Testing Requirements

Unit Tests

Dependencies

Required Packages

No External Dependencies

Acceptance Criteria

Best Practices

Good Tool Docstrings

Clear Parameter Names

Consistent Return Types

Limitations

No Advanced Type Hints

No Parameter Validation

Future Enhancements