Skip to main content

Minimal Python function calling for Claude

Project description

toololo

Minimal Python function calling for Claude

logo

Toololo is a tiny library for using Python functions as tools in Claude. It does two things:

  • Automatically creates tool use schemas for provided functions
  • Implements a Think/Write/Call loop

Usage

The following code will run until Claude considers itself done with the task, or max_iterations are exhausted:

import toololo
import anthropic

client = anthropic.Client()

generator = toololo.run(
    client=client,
    messages=messages,   # str or list[dict]
    model=claude_model,  # e.g. "claude-3-7-sonnet-latest
    tools=list_of_functions,
    system_prompt=system_prompt,
    max_tokens=8192,
    thinking_budget=4096,
    max_iterations=50,
)
for output in generator:
    print(output)

output is one of ThinkingContent, TextContent, ToolUseContent, ToolResult (types are defined in types.py).

Examples

Call Python functions

Give Claude access to arbitrary Python functions:

import subprocess
import anthropic
import toololo

def curl(args: list[str]) -> str:
    if "-m" not in args and "--max-time" not in args:
        args = args + ["--max-time", "30"]

    result = subprocess.run(["curl"] + args, capture_output=True, text=True, timeout=60)
    if result.returncode != 0:
        return f"Error (code {result.returncode}): {result.stderr}"

    return result.stdout

if __name__ == "__main__":
    client = anthropic.Client()

    prompt = "Do a basic network speed test and analyze the results."

    for output in toololo.run(
        client,
        prompt,
        model="claude-3-7-sonnet-latest",
        tools=[curl],
    ):
        print(output)

Call methods on objects

You can also call methods on objects with state:

import anthropic
import toololo

class TowersOfHanoi:
    def __init__(self, num_disks=3):
        self.towers = [list(range(num_disks, 0, -1)), [], []]
        self.num_disks = num_disks

    def get_state(self) -> list[list[int]]:
        return self.towers

    def move(self, from_index: int, to_index: int) -> None:
        if not (0 <= from_index <= 2 and 0 <= to_index <= 2):
            raise ValueError("Tower index must be 0, 1, or 2")

        if not self.towers[from_index]:
            raise ValueError(f"Cannot move from empty tower {from_index}")

        if (
            self.towers[to_index] and self.towers[from_index][-1] > self.towers[to_index][-1]
        ):
            raise ValueError("Cannot place larger disk on top of smaller disk")

        disk = self.towers[from_index].pop()
        self.towers[to_index].append(disk)

    def is_complete(self) -> bool:
        return len(self.towers[2]) == self.num_disks

if __name__ == "__main__":
    client = anthropic.Client()
    towers = TowersOfHanoi()

    assert not towers.is_complete()

    for output in toololo.run(
        client,
        messages=[
            {
                "role": "user",
                "content": "Solve this Towers of Hanoi puzzle. The goal is to move all disks from the first tower (index 0) to the third tower (index 2). You can only move one disk at a time, and you cannot place a larger disk on top of a smaller disk.",
            }
        ],
        model="claude-3-7-sonnet-latest",
        tools=[towers.get_state, towers.move, towers.is_complete],
    ):
        print(output)

    assert towers.is_complete()

Multi-agent system

By instantiating two toololo.run generators, we can create cooperating or competitive multi-agent systems.

import anthropic
import toololo

class TicTacToe:
    def __init__(self):
        self.board: list[list[str | None]] = [
            [None for _ in range(3)] for _ in range(3)
        ]
        self.current_player = "X"
        self.winner = None
        self.game_over = False

    def get_board(self) -> list[list[str | None]]:
        return self.board

    def is_game_over(self) -> bool:
        return self.game_over

    def get_winner(self) -> str | None:
        return self.winner

    def make_move(self, row: int, col: int) -> bool:
        # Validate move
        if (
            self.game_over
            or not (0 <= row < 3 and 0 <= col < 3)
            or self.board[row][col] is not None
        ):
            return False

        # Make the move
        self.board[row][col] = self.current_player

        # Check for win
        if self.check_win():
            self.winner = self.current_player
            self.game_over = True
        # Check for draw
        elif all(self.board[r][c] is not None for r in range(3) for c in range(3)):
            self.game_over = True

        # Switch player
        self.current_player = "O" if self.current_player == "X" else "X"
        return True

    def check_win(self) -> bool:
        p = self.current_player
        b = self.board

        # Check rows, columns and diagonals
        for i in range(3):
            if (
                b[i][0] == b[i][1] == b[i][2] == p  # rows
                or b[0][i] == b[1][i] == b[2][i] == p  # columns
            ):
                return True

        return (
            b[0][0] == b[1][1] == b[2][2] == p  # diagonal
            or b[0][2] == b[1][1] == b[2][0] == p  # diagonal
        )

    def print_board(self) -> None:
        for i, row in enumerate(self.board):
            print(" | ".join([cell if cell else " " for cell in row]))
            if i < len(self.board) - 1:
                print("-" * 9)


if __name__ == "__main__":
    client = anthropic.Client()
    game = TicTacToe()

    x_prompt = "You are player X"
    o_prompt = "You are player O"
    system_prompt = "You're playing a game of Tic-Tac-Toe. The other player will automatically make moves in between your moves. Keep playing until there's a winner or a draw"

    print("=== Starting Tic-Tac-Toe Game ===")
    game.print_board()

    tools = [
        game.get_board,
        game.make_move,
        game.is_game_over,
        game.get_winner,
    ]

    x_generator = toololo.run(
        client,
        messages=x_prompt,
        model="claude-3-7-sonnet-latest",
        tools=tools,
        system_prompt=system_prompt,
    )

    o_generator = toololo.run(
        client,
        messages=o_prompt,
        model="claude-3-7-sonnet-latest",
        tools=tools,
        system_prompt=system_prompt,
    )

    while not game.is_game_over():
        current_player = game.current_player
        current_gen = x_generator if current_player == "X" else o_generator

        output = next(current_gen)

        if isinstance(output, toololo.types.ToolResult):
            if output.func == game.make_move and output.success:
                print("\nCurrent board:")
                game.print_board()

    print("\n=== Game Over ===")
    game.print_board()

    winner = game.get_winner()
    if winner:
        print(f"Player {winner} wins!")
    else:
        print("It's a draw!")

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

toololo-0.1.1.tar.gz (13.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

toololo-0.1.1-py3-none-any.whl (12.8 kB view details)

Uploaded Python 3

File details

Details for the file toololo-0.1.1.tar.gz.

File metadata

  • Download URL: toololo-0.1.1.tar.gz
  • Upload date:
  • Size: 13.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.12.7

File hashes

Hashes for toololo-0.1.1.tar.gz
Algorithm Hash digest
SHA256 0f41f7241fe2519ea2d847d2fa98f3c2cf7fbeb2181b2e9572211dd48a0ed957
MD5 b3d5c2ddc19e06e8bff04991c545de45
BLAKE2b-256 699f6c88f9135a8f430d62eb469b8d2a16aa1b51ee03e2c63d3d9b641bedbedb

See more details on using hashes here.

File details

Details for the file toololo-0.1.1-py3-none-any.whl.

File metadata

  • Download URL: toololo-0.1.1-py3-none-any.whl
  • Upload date:
  • Size: 12.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.12.7

File hashes

Hashes for toololo-0.1.1-py3-none-any.whl
Algorithm Hash digest
SHA256 2572c9c9eb93246844c243eeeb3898bd17d549cfb5ea5748777c81892e17df6e
MD5 0dfa1d223447a58be1f355f4526ff3c1
BLAKE2b-256 f6517828a009ff91315514d8d7f66ccb592456246bea9dffc58cb5c31a3d894e

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page