This repository is a curated list of language agents, categorized by their underlying cognitive architectures and functionalities, based on the "Cognitive Architectures for Language Agents" paper. It serves researchers and developers in the LLM agent space by providing a structured overview of existing work, enabling easier comparison and identification of relevant approaches for building more capable and grounded AI agents.

How It Works

The CoALA framework defines language agents by their action spaces: external actions for environment interaction (grounding) and internal actions for memory interaction (reasoning, retrieval, learning). Agents operate in decision-making cycles comprising planning (proposing and evaluating actions using internal reasoning/retrieval) and execution (performing selected actions to update memory or the environment). This structured approach allows for a systematic classification and understanding of how different agents manage memory, reasoning, and interaction.

Quick Start & Requirements

This repository is a list of papers and does not have a direct installation or execution command. It requires no specific software to view.

Highlighted Details

• Comprehensive list of over 100 language agent papers with categorized actions (grounding, reasoning, retrieval, learning).
• Links to the foundational CoALA paper, related tweets, and a BibTeX file with over 300 citations.
• Includes a variety of environments and benchmarks for evaluating agent performance.
• Actively seeks contributions via pull requests to expand the list.

Maintenance & Community

The project is maintained by ysymyth and welcomes community contributions. Links to related resources like "LLM Powered Autonomous Agents" and "awesome-llm-powered-agent" are provided.

Licensing & Compatibility

The repository itself is not licensed, but it references academic papers and frameworks. Users should consult the licenses of individual projects linked within the list.

Limitations & Caveats

The list is a subset of existing work and may contain incorrect action space labels due to scraping and pull requests. The project plans to add more papers and labels over time.