Autochat

⚠️ Warning: Since agentic capabilities are evolving fast, expect the API to change.

A lightweight Python library to build AI agents with LLMs.

Key Features

• 🤝 Support for multiple LLM providers (OpenAI and Anthropic)
• 🐍 Transform python function or class into a tool
• 🔁 Run conversation as a generator.
• 🙈 Handle caching by default (anthropic model claude-3-7-sonnet-latest)
• ✨ And more features including:
  • Simple template system
  • Easy function and tool integration
  • Flexible instruction and example management
  • Support for images
  • Support for MCP servers

Example (search capability)

The library supports function call, handling the back-and-forth between the system and the assistant.

from autochat import Autochat

def search_wikipedia(title: str):
    """Search wikipedia for information"""
    import requests
    from bs4 import BeautifulSoup

    response = requests.get(f"https://en.wikipedia.org/w/index.php?search={title}&title=Special%3ASearch")
    soup = BeautifulSoup(response.text, 'html.parser')
    body_content = soup.find('div', {'id': 'bodyContent'})
    return body_content.text.strip()

classifier_agent = Autochat()
classifier_agent.add_function(search_wikipedia)

text = "Since when is the lastest iphone available?"
for message in classifier_agent.run_conversation(text):
    print(message.to_markdown())

# > ## user
# > Since when is the lastest iphone available?
# > ## assistant
# > search_wikipedia(title=iPhone)
# > ## function
# > Result: (html content)
# > ## assistant
# > The latest flagship iPhone models, the iPhone 16 and 16 Plus, along with the higher-end iPhone 16 Pro and 16 Pro Max, were available as of January 1, 2024.

Quick Start

Initialize with OpenAI (default)

agent = Autochat(instruction="You are a helpful assistant")

Simple conversation

response = agent.ask("What is the capital of France?")
print(response.content)

Using Anthropic's Claude

agent = Autochat(provider="anthropic")
response = agent.ask("Explain quantum computing in simple terms")
print(response.content)

Run conversation as a generator

for message in agent.run_conversation("Explain quantum computing in simple terms"):
    print(message.to_markdown())

Async Interface

Autochat provides async versions of its core methods for use in async applications:

# Async version of ask
response = await agent.ask_async("What is the capital of France?")
print(response.content)

# Async version of run_conversation
async for message in agent.run_conversation_async("Explain quantum computing"):
    print(message.to_markdown())

# Async function calls are also supported
async def async_calculator(a: int, b: int) -> int:
    await asyncio.sleep(0.1)  # Some async work
    return a + b

agent.add_function(async_calculator)
async for message in agent.run_conversation_async("What is 5 + 3?"):
    print(message.to_markdown())

Add a function call as python function

def multiply(a: int, b: int) -> int:
    return a * b

agent = Autochat()
agent.add_function(multiply)
text = "What is 343354 * 13243343214"
for message in agent.run_conversation(text):
    print(message.to_markdown())

Add a Class as a tool

from autochat import Autochat

class Calculator:
    def add(self, a: int, b: int) -> int:
        """Add two numbers"""
        return a + b

    def multiply(self, a: int, b: int) -> int:
        """Multiply two numbers"""
        return a * b

calculator = Calculator()

agent = Autochat()
agent.add_tool(calculator)
for message in agent.run_conversation(
    "What make 343354 * 13243343214"
):
    print(message)

Add a MCP server

Experimental feature. Check out the tests/mcp_clients for more information.

Installation

To install the package, you can use pip:

pip install 'autochat[all]'

Image support

from autochat import Autochat, Message

from PIL import Image

agent = Autochat()

image = Image.open("examples/image.jpg")
message = Message(role="user", content="describe the image", image=image)
response = agent.ask(message)
print(response.to_markdown())

Template System

We provide a simple template system for defining the behavior of the chatbot, using markdown-like syntax.

## system
You are a parrot

## user
Hi my name is Bob

## assistant
Hi my name is Bob, hi my name is Bob!

## user
Can you tell me my name?

## assistant
Your name is Bob, your name is Bob!

You can then load the template file using the from_template method:

parrotGPT = Autochat.from_template("./parrot_template.txt")

The template system also supports function calls. Check out the examples/demo_label.py for a complete example.

Environment Variables

The AUTOCHAT_MODEL environment variable specifies the model to use. If not set, it defaults to "gpt-4o" for openai and "claude-3-7-sonnet-latest" for anthropic.

We recommend to use Anthropic / claude-3-7-sonnet-latest for agentic behavior.

export AUTOCHAT_MODEL="gpt-4o"
export OPENAI_API_KEY=<your-key>

or with anthropic

export AUTOCHAT_MODEL="claude-3-7-sonnet-latest"
export ANTHROPIC_API_KEY=<your-key>

Use AUTOCHAT_HOST to use alternative provider (openai, anthropic, openpipe, llama_cpp, ...)

Support

If you encounter any issues or have questions, please file an issue on the GitHub project page.

License

This project is licensed under the terms of the MIT license.