LiteAgent is an easy-to-learn, lightweight, and extensible AI agent framework built on top of LiteLLM. It is designed as a minimal yet practical implementation for quickly building intelligent assistants and chatbots with robust tool-calling capabilities. The codebase is intentionally simple, making it ideal for learning, extension, and rapid prototyping.

Key Advantages:

• Minimal and approachable: The simplest agent implementation for fast learning and hacking.
• Accurate and complete type hints: All function signatures are fully type-hinted and never faked, ensuring reliable developer experience and static analysis.
• Flexible parameter definition: Supports defining tool function parameters using basic types, Pydantic models, or Python dataclasses—even in combination.
• Streaming responses: Seamless support for LiteLLM streaming output.
• Custom tool functions: Easily integrate your own Python functions (e.g., weather, temperature queries).
• Rich type annotations, Pydantic-based.
• Easy to extend and test.

You can install LiteAgent directly from PyPI:

pip install lite-agent

Or use uv:

uv pip install lite-agent

If you want to install from source for development:

uv pip install -e .
# or
pip install -e .

See examples/basic.py:

import asyncio
from lite_agent.agent import Agent
from lite_agent.runner import Runner
async def get_whether(city: str) -> str:
    await asyncio.sleep(1)
    return f"The weather in {city} is sunny with a few clouds."
async def main():
    agent = Agent(
        model="gpt-4.1",
        name="Weather Assistant",
        instructions="You are a helpful weather assistant.",
        tools=[get_whether],
    )
    runner = Runner(agent)
    resp = await runner.run_until_complete("What's the weather in New York?")
    for chunk in resp:
        print(chunk)
if __name__ == "__main__":
    asyncio.run(main())

See pyproject.toml for details.

pytest

MIT License

Jianqi Pan (jannchie@gmail.com)