Skip to main content

Minimal Python function calling for Claude

Project description

toololo

PyPI - Version

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

Install

pip install toololo

Usage

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

import asyncio
import toololo
import anthropic

async def main():
    client = anthropic.AsyncClient()

    async for output in 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,
    ):
        print(output)

asyncio.run(main())

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 asyncio
import subprocess
import anthropic
import toololo

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

    process = await asyncio.create_subprocess_exec(
        "curl",
        *args,
        stdout=asyncio.subprocess.PIPE,
        stderr=asyncio.subprocess.PIPE
    )

    stdout, stderr = await process.communicate()

    if process.returncode != 0:
        return f"Error (code {process.returncode}): {stderr.decode()}"

    return stdout.decode()

async def main():
    client = anthropic.AsyncClient()

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

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

if __name__ == "__main__":
    asyncio.run(main())

Interactive prompting

import asyncio
import toololo

async def main():
    tools = [...]  # TODO: fill out with your own tools
    messages = []

    while True:
        print("> ", end="")
        prompt = input()
        if prompt.lower() in ["q", "quit", "exit"]:
            return

        run = toololo.Run(
            client=anthropic_client,
            model="claude-3-7-sonnet-latest",
            messages=messages + [{"role": "user", "content": prompt}],
            tools=tools,
            max_iterations=200,
        )
        try:
            async for output in run:
                if not isinstance(output, toololo.types.ToolResult):
                    print(output)
                    print()
        except (KeyboardInterrupt, asyncio.exceptions.CancelledError):
            print("\n[Generation interrupted]")

        messages = run.messages


if __name__ == "__main__":
    asyncio.run(main(prompt))

Call methods on objects

You can also call methods on objects with state:

import asyncio
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

async def main():
    client = anthropic.AsyncClient()
    towers = TowersOfHanoi()

    assert not towers.is_complete()

    async 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()

if __name__ == "__main__":
    asyncio.run(main())

Multi-agent system

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

import asyncio
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)

async def main():
    client = anthropic.AsyncClient()
    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,
    ]

    def create_generator(prompt):
        return toololo.Run(
            client,
            messages=prompt,
            model="claude-3-7-sonnet-latest",
            tools=tools,
            system_prompt=system_prompt,
        )

    x_generator = create_generator(x_prompt)
    o_generator = create_generator(o_prompt)

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

        try:
            output = await anext(current_gen)
        except StopAsyncIteration:
            # Reinitialize the stopped generator
            if current_player == "X":
                x_generator = create_generator(x_prompt)
                current_gen = x_generator
            else:
                o_generator = create_generator(o_prompt)
                current_gen = o_generator
            output = await anext(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!")

if __name__ == "__main__":
    asyncio.run(main())

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-0.4.1.tar.gz (15.9 kB view details)

Uploaded Source

Built Distribution

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

toololo-0.4.1-py3-none-any.whl (15.0 kB view details)

Uploaded Python 3

File details

Details for the file toololo-0.4.1.tar.gz.

File metadata

  • Download URL: toololo-0.4.1.tar.gz
  • Upload date:
  • Size: 15.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.12.9

File hashes

Hashes for toololo-0.4.1.tar.gz
Algorithm Hash digest
SHA256 342ee3ec2b9ba834a3a7983c77fc3f426f64c84521a1f4f4e9569b3d34bbc415
MD5 bf862e214cd8944b83ed0bd8dbd3506b
BLAKE2b-256 5132259fa37a370b26b3bf33963b3e2d7d3fed31d48cabe8b16a399f3013ed73

See more details on using hashes here.

Provenance

The following attestation bundles were made for toololo-0.4.1.tar.gz:

Publisher: ci.yaml on andreasjansson/toololo

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file toololo-0.4.1-py3-none-any.whl.

File metadata

  • Download URL: toololo-0.4.1-py3-none-any.whl
  • Upload date:
  • Size: 15.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.12.9

File hashes

Hashes for toololo-0.4.1-py3-none-any.whl
Algorithm Hash digest
SHA256 9048992daec77767c571f5c9e8bbd2dd2bd718ca873e103c6d2324ba9e57d0ed
MD5 44acff3abc197858077f0961d1063196
BLAKE2b-256 9bb1db5d6f1580575c5e5182e4824370c353b7e33e4a972d15421800b0836569

See more details on using hashes here.

Provenance

The following attestation bundles were made for toololo-0.4.1-py3-none-any.whl:

Publisher: ci.yaml on andreasjansson/toololo

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

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