I built AgentSilex because I was frustrated trying to debug and customize existing agent frameworks. When something went wrong, I'd spend hours digging through thousands of lines of abstraction layers.

AgentSilex is different: the core is ~300 lines of readable Python. You can actually understand how your agents work.

Quick example - a working agent in 20 lines:

    from agentsilex import Agent, Runner, Session, tool

    @tool
    def get_weather(city: str) -> str:
        return "SUNNY"

    agent = Agent(
        name="Assistant",
        model="gpt-4o-mini",  # or any LLM via LiteLLM
        instructions="Help with weather info",
        tools=[get_weather]
    )

    result = Runner(Session()).run(agent, "What's the weather in Paris?")

It's designed for: - Developers who want to understand agent internals - Companies needing a customizable foundation - Anyone learning how LLM agents actually work

The trade-off: minimal built-in tools. You'll write more glue code, but you'll know exactly what your system is doing.

I'd love feedback on: - Is the API intuitive? - What use cases am I missing? - Would you actually use this over existing frameworks?

pip install agentsilex

GitHub: https://github.com/howl-anderson/agentsilex

Looking forward to your thoughts!