llmadapt

A small, zero-dependency, provider-agnostic agent for chatting with LLMs and giving them tools. Works with Anthropic, OpenAI, Gemini, and Ollama using only the Python standard library for HTTP — no requests, no provider SDKs. v0.1.0

Install

From source (until published to PyPI):

git clone https://github.com/Shyam-Parikh-2025/llmadapt.git
cd llmadapt
pip install -e .

Once published:

pip install llmadapt

Quick start

from llmadapt import Agent

agent = Agent(
    provider="anthropic",
    model="claude-sonnet-4-6",
    api_key="sk-ant-...",  # or set ANTHROPIC_API_KEY in your environment
    system_instruction="You are a helpful assistant.",
)

print(agent.chat("What's 12 * 7?"))

Giving the agent tools

def get_weather(city: str) -> dict:
    """Look up the current weather for a city."""
    return {"city": city, "tempF": 72, "condition": "sunny"}

agent.add_tool(get_weather)
print(agent.chat("What's the weather in NYC?"))

A JSON schema is auto-generated from the function's type hints, default values, and docstring. Tool outputs are JSON-serialized automatically (plain strings pass through untouched).

Switching providers

agent.switch_api(provider="openai", model="gpt-4o", api_key="sk-...")

Limiting tool-call loops

agent.set_max_tool_iterations(20)  # default is 10

If the model gets stuck calling tools repeatedly without producing a final answer, chat() raises a RuntimeError instead of looping forever.

Supported providers

Running tests

python tests/test_agent.py

Tests run entirely offline against scripted fake HTTP responses — no API key or network access required.

License

MIT