Skip to content

Welcome to LLLM

LLLM Logo

LLLM is a lightweight framework designed for researchers to fast-prototype advanced agentic systems. It emphasizes minimalism, modularity, and type safety, with first-class support for program synthesis and neuro-symbolic capabilities.

Key Features

  • Modular Architecture: Core abstractions, providers, tools, and memory are decoupled.
  • Type Safety: Built on Pydantic for robust data validation and strict typing.
  • Provider Interface: First-class OpenAI support with an extensible interface for adding more providers.
  • Neuro-Symbolic Design: Advanced prompt management with structured output, exception handling, and interrupt logic.
  • API Proxies: Secure code execution of external APIs for program synthesis.

Quick Start

pip install lllm

See Getting Started for more details.

Experimental Features

  • lllm.tools.cua (Computer Use Agent) integrates with Playwright and the OpenAI Computer Use API. Expect rapid iteration and breaking changes.
  • Responses API routing is available per agent via api_type = "response"; today it only targets OpenAI models that advertise the requisite features.

Work in Progress

  • Additional providers (Anthropic, Gemini, local runtimes) are on the roadmap but not yet available in the lllm.providers registry.
  • Provider-agnostic streaming and richer auto-discovery ergonomics (hot-reloading prompts/proxies) are actively being designed.