OpenAgentKit

A comprehensive open-source toolkit for building agentic applications. OpenAgentKit provides a unified interface to work with various LLM providers, tools, and agent frameworks.

WARNING: Everything here is still in development, expect many bugs and unsupported features, please feel free to contribute!

Features

• Lightweight Structure: Keeping core features of AI agents while still create rooms for custom extension without cluttering.
• Unified LLM Interface: Consistent API across multiple LLM providers by leveraging OpenAI APIs (will be extended in the future!)
• Generator-based event stream: Event-driven processing using a generator
• Async Support: Built-in asynchronous processing for high-performance applications
• Tool Integration: Pre-built tools for common agent tasks
• Extensible Architecture: Easily add custom models and tools
• Type Safety: Comprehensive typing support with Pydantic models

Installation

pip install openagentkit==0.1.0a14

Quick Start

from openagentkit.modules.openai import OpenAIAgent
from openagentkit.core.tools.base_tool import tool
from pydantic import BaseModel
import openai
import os
import json

# Define a tool
@tool # Wrap the function in a tool decorator to automatically create a schema
def get_weather(city: str):
    """Get the weather of a city"""

    # Actual implementation here...
    # ...

    return f"Weather in {city}: sunny, 20°C, feels like 22°C, humidity: 50%"

# Initialize OpenAI client
client = openai.OpenAI(
    api_key=os.getenv("OPENAI_API_KEY"),
)

agent = OpenAIAgent(
    client=client,
    model="gpt-4o-mini",
    system_message="""
    You are a helpful assistant that can answer questions and help with tasks.
    You are also able to use tools to get information.
    """,
    tools=[get_weather],
    temperature=0.5,
    max_tokens=100,
    top_p=1.0,
)

generator = agent.execute(
    messages=[
        {"role": "user", "content": "What's the weather like in New York?"}
    ],
)

for response in generator:
    print(response)

print(json.dumps(agent.get_history(), indent=2))

Supported Integrations

• LLM Providers:

  • OpenAI
  • SmallestAI
  • Azure OpenAI (via OpenAI integration)
  • More coming soon!

• Tools (Mostly for prototyping purposes):

  • Weather information (Requires WEATHERAPI_API_KEY)

Architecture

OpenAgentKit is built with a modular architecture:

• Interfaces: Abstract base classes defining the contract for all implementations
• Models: Pydantic models for type-safe data handling
• Modules: Implementation of various services and integrations
• Handlers: Processors for tools and other extensions
• Utils: Helper functions and utilities

Advanced Usage

Asynchronous Processing

from openagentkit.modules.openai import OpenAIAgent
from openagentkit.core.tools.base_tool import tool
from pydantic import BaseModel
from typing import Annotated
import asyncio
import openai
import os

# Define an async tool
@tool # Wrap the function in a tool decorator to automatically create a schema
async def get_weather(city: str):
    """Get the weather of a city"""

    # Actual implementation here...
    # ...

    return f"Weather in {city}: sunny, 20°C, feels like 22°C, humidity: 50%"

# Initialize OpenAI client
client = openai.AsyncOpenAI(
    api_key=os.getenv("OPENAI_API_KEY"),
)

async def main():
    # Initialize LLM service
    agent = AsyncOpenAIAgent(
        client=client,
        model="gpt-4o-mini",
        system_message="""
        You are a helpful assistant that can answer questions and help with tasks.
        You are also able to use tools to get information.
        """,
        tools=[get_weather],
        temperature=0.5,
        max_tokens=100,
        top_p=1.0,
    )

    generator = agent.execute(
        messages=[
            {"role": "user", "content": "What's the weather like in New York?"}
        ]
    )

    async for response in generator:
        print(response.content)

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

Custom Tool Integration

Using the @tool decorator:

from openagentkit.core.utils.tool_wrapper import tool
from pydantic import BaseModel
from typing import Annotated

# Define a tool
@tool # Wrap the function in a tool decorator to automatically create a schema
def get_weather(city: str):
    """Get the weather of a city""" # Always try to add pydoc in the function for better comprehension by LLM 

    # Actual implementation here...
    # ...

    return f"Weather in {city}: sunny, 20°C, feels like 22°C, humidity: 50%"

# Get the tool schema
print(get_weather.schema)

# Run the tool like any other function
weather_response = get_weather("Hanoi")
print(weather_response) 

By subclassing Tool:

from openagentkit.core.tools.base_tool import Tool

class GetWeather(Tool):
    """
    A tool to get the current weather of a city.
    """
    def __call__(self, city: str) -> str:
        """
        Get the current weather in a city.
        """
        # Simulate a weather API call
        return f"The current weather in {city} is sunny with a temperature of 25°C."
    
get_weather = GetWeather()

# Get the tool schema
print(get_weather.schema)

# Run the tool like any other function
weather_response = get_weather("Hanoi")
print(weather_response) 

Custom Context Store

An Agent must have access to context (chat) history to be truly an agent. OpenAgentKit has a ContextStore module that supports various cache providers (Redis, Valkey) and a quick module for testing (InMemory)

from openagentkit.modules.openai import AsyncOpenAIAgent
from openagentkit.core.context import InMemoryContextStore
import openai
import asyncio
from dotenv import load_dotenv
from pydantic import BaseModel
import os

load_dotenv()

context_store = InMemoryContextStore()

async def main():
    client = openai.AsyncOpenAI(
        api_key=os.getenv("OPENAI_API_KEY"),
    )

    # When initializing an agent, you can pass in a thread_id or agent_id as identifier for the default context scope. The 2 values are immutable for consistency.
    agent = AsyncOpenAIAgent(
        client=client,
        system_message="You are a helpful assistant.",
        context_store=context_store,
        thread_id="test"
        agent_id="AssistantA"
    )

    # Access the thread_id property
    print(f"Thread ID: {agent.thread_id}")

    async for event in agent.execute(
        messages=[
            {
                "role": "user",
                "content": "Hi, my name is John."
            }
        ]
    ):
        if event.content:
            print(f"Response: {event.content}")

    # If no thread_id is defined when executing the agent, it will defaults to the initialized thread_id attribute.
    async for event in agent.execute(
        messages=[
            {
                "role": "user",
                "content": "What is my name?"
            }
        ]
    ):
        if event.content:
            print(f"Response: {event.content}")

    async for event in agent.execute(
        messages=[
            {
                "role": "user",
                "content": "What is my name?"
            }
        ],
        thread_id="new_context" # Since this is a new thread, the agent will no longer knowledge of the previous interaction
    ):
        if event.content:
            print(f"Response: {event.content}")

    # If no thread_id is defined when executing the agent, it will defaults to the initialized thread_id attribute.
    async for event in agent.execute(
        messages=[
            {
                "role": "user",
                "content": "Okay lovely, can you refer me to my name at the end of your sentence always?"
            }
        ],
    ):
        if event.content:
            print(f"Response: {event.content}")

    # Get Contexts related to agent instance (Agent ID)
    print(context_store.get_agent_context(agent.agent_id))

if __name__ == "__main__":
    asyncio.run(main())
    # Get Context from thread_id
    print(context_store.get_context("new_context"))

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.