Awesome AI Agents › Multi-Agent Frameworks

samholt/L2MAC

⭐ 160 Python added to this list on 2025-04-19 repository created 2024-03-08

L2MAC (Large Language Model Automatic Computer) is a pioneering framework that implements the first practical LLM-based general-purpose stored-program automatic computer following the von Neumann architecture. It is designed as an LLM-based multi-agent system capable of solving large and complex tasks by generating extensive and consistent outputs that are not limited by the fixed context window constraints of traditional large language models (LLMs). The framework excels in generating large codebases, writing entire books, and other tasks requiring unbounded output generation from a single input prompt. L2MAC operates by internally using a "prompt-program," which is a sequence of instruction steps executed by individual LLM agents. Each step is loaded into a new LLM agent whose context is managed by a control unit. These agents are equipped with tools to read and write to a persistent memory, such as a file store, which holds both intermediate and final outputs. This architecture allows L2MAC to automatically execute general-purpose prompt programs to solve complex tasks requiring extensive cohesive outputs. The framework supports generating complex codebases from a single user prompt, with additional tools to check for syntax errors and run unit tests. It is open-source, actively developed, and presented at major conferences like ICLR 2024. L2MAC can be used via a command-line interface or as a Python library, making it accessible for various use cases. The project provides comprehensive documentation, tutorials, and examples to help users get started and explore its capabilities. L2MAC represents a significant advancement in leveraging LLMs for large-scale automated code generation and complex task solving, overcoming traditional limitations of LLM context windows by employing a multi-agent system and persistent memory management. It is licensed under MIT and encourages community contributions and engagement through Discord and GitHub.

https://github.com/samholt/L2MAC

agentautomated-code-generationautomatic-computerbookcodebasecodebase-generationcodingcommand-line-interfacecomplex-task-solvingextensive-output-generationfile-storegeneratorgpticlr-2024l2maclarge-language-modelllmmulti-agentmulti-agent-systemopen-sourcepersistent-memoryprompt-programpython-librarysyntax-error-checkingunbounded-outputunit-testingvon-neumann-architecture