GitHub
FreeThe LLM Automatic Computer Framework
FreeFree tier
Inputs: textOutputs: code, text
About GitHub
L2MAC (Large Language Model Automatic Computer) is a pioneering framework that implements a stored-program automatic computer (von Neumann architecture) using LLM-based multi-agent systems. It solves complex tasks requiring extensive and consistent outputs, bypassing the fixed context window constraint of LLMs. L2MAC uses a self-generated prompt-program composed of instruction steps, each executed by a new LLM agent. A control unit manages context and provides tools for reading/writing to a persistent file store. This enables automatic execution of general-purpose prompt programs to generate large codebases, entire books, and other cohesive outputs.
Key Features
Implements a stored-program automatic computer (von Neumann architecture) using LLM agents
Self-generated prompt-program bootstrapped from a single user prompt
Each instruction step executed by a separate LLM agent with managed context
Persistent file store memory enables unbounded output generation
Multi-agent system with control unit for context management and tool provision
State-of-the-art for generating large codebases and entire books
Bypasses fixed context window constraints of standard LLMs
Pros & Cons
Pros
- Pioneering practical LLM-based general-purpose stored-program computer
- State-of-the-art performance for large code generation tasks
- Accepted and presented at ICLR 2024
- Fully open-source (MIT license via GitHub)
- Unbounded output not limited by context window
- Easy installation via pip
Cons
- Requires Python 3.7+ and an LLM backend (e.g., OpenAI API) to function
- Complex internal architecture may have a learning curve for configuration
- Output quality heavily depends on underlying LLM model used
Best For
Generating large complex codebases for entire applications from a single promptWriting extensive books or documents requiring cohesive long-form contentSolving complex tasks that demand extensive and consistent multi-step outputsAutomating pipeline tasks requiring sequential instruction execution by LLMs
FAQ
What is L2MAC?
L2MAC (Large Language Model Automatic Computer) is an open-source framework that implements a stored-program automatic computer using LLM agents. It can generate extensive and consistent outputs, such as entire codebases or books, without being constrained by the LLM's fixed context window.
How do I install L2MAC?
Installation requires Python 3.7+. Use `pip install --upgrade l2mac` or clone the repository and install with `pip install --upgrade -e .`.
Does L2MAC require an API key?
Yes, L2MAC relies on an underlying LLM (e.g., GPT-4). You must configure the API key in the config file as described in the documentation.
What makes L2MAC different from other LLM agents?
L2MAC uses a von Neumann-style architecture with a stored program and persistent memory, allowing it to execute multi-step prompt programs that generate outputs unlimited by context window constraints.