Quickstart

Install

uv add lexigram-ai-llm
# Or with pip
pip install lexigram-ai-llm

Install a provider extra (at least one is required):

uv add "lexigram-ai-llm[openai]"
# or: uv add "lexigram-ai-llm[anthropic]"
# or: uv add "lexigram-ai-llm[ollama]"

Minimal LLM call

import asyncio
from lexigram import Application
from lexigram.ai.llm import LLMModule, ClientConfig


async def main() -> None:
    config = ClientConfig(
        provider="openai",
        model="gpt-4o",
        api_key="sk-...",
    )

    async with Application.boot(
        name="llm-demo",
        modules=[LLMModule.configure(config)],
    ) as app:
        from lexigram.contracts.ai import LLMClientProtocol

        llm = await app.container.resolve(LLMClientProtocol)
        result = await llm.complete([
            {"role": "user", "content": "What is the capital of France?"},
        ])

        if result.is_ok():
            print(result.unwrap().content)  # "Paris"
        else:
            print(f"Error: {result.unwrap_err()}")


asyncio.run(main())

How it works

LLMModule.configure(config) creates a DynamicModule with an LLMProvider
The provider creates a client via create_llm_client() using the ProviderRegistry
LLMClientProtocol is registered in the container — resolve it anywhere
llm.complete() returns Result[CompletionProtocol, LLMError]

Next steps

Guide — mental model, conversations, streaming, structured output
Architecture — client layers, cache, model management
Configuration — every config key
How-tos — common recipes