llmini

A minimal, ergonomic Python library for interacting with LLMs. Built on top of LiteLLM.

Features

• Unified LLM access — works with Anthropic, OpenAI, and any LiteLLM-supported provider
• Tool calling — define tools as Python functions, schemas are auto-generated from type hints and docstrings
• Structured output — parse LLM responses into Pydantic models
• Image support — send images in messages with automatic base64 encoding
• Approval gating — mark tools as requiring approval before execution
• Message serialization — export conversations as JSON or JSONL
• Zero config — works out of the box with environment variables for API keys

Installation

uv add llmini

# or
pip install llmini

Quick Start

from llmini import ModelConfig, complete, system_message, user_message

reply = complete(
    model=ModelConfig(model="anthropic/claude-3-5-haiku-latest"),
    messages=[
        system_message("You are a helpful assistant."),
        user_message("What is the capital of France?"),
    ],
)

print(reply.content)  # "The capital of France is Paris."

With tools:

from llmini import ModelConfig, complete, system_message, user_message, make_tool

@make_tool()
def get_weather(location: str) -> str:
    """Get the current weather at a location."""
    return "Sunny, 22°C"

reply = complete(
    model=ModelConfig(model="anthropic/claude-3-5-haiku-latest"),
    messages=[
        system_message("You are a helpful assistant."),
        user_message("How's the weather in Paris?"),
    ],
    tools=[get_weather],
)

Building a tool-using agent loop: see examples/test_04_agent.py.

Documentation

• USAGE.md — full API reference with all parameters, defaults, and detailed examples
• examples/ — runnable example scripts

License

MIT