# Cognitive Architectures for Language Agents

**Authors:** Theodore R. Sumers, Shunyu Yao, Karthik Narasimhan, Thomas L. Griffiths  
**arXiv:** [2309.02427](https://arxiv.org/abs/2309.02427)  
**Date:** 2023-09-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2309.02427v2/

## Key points

- Recent advancements in language models have led to the development of a new class of language agents
- These agents combine large language models (LLMs) with external resources or internal control flows
- They have shown success in tasks requiring grounding or reasoning
- Currently, there is no systematic framework to organize these agents and plan for future developments
- The authors propose Cognitive Architectures for Language Agents (CoALA) as a solution
- CoALA describes a language agent with modular memory components, structured action space, and generalized decision-making process
- Using CoALA, the authors retrospectively survey and organize recent work in this field
- They also identify actionable directions for developing more capable language agents
- CoALA framework contextualizes today's language agents within the broader history of AI
- It outlines a path towards achieving language-based general intelligence
- This comprehensive framework provides a foundation for organizing existing research efforts and guiding future developments in the field of language agents.

## Summary

Recent advancements in language models have resulted in the development of a new class of language agents that combine large language models (LLMs) with external resources or internal control flows. These agents have demonstrated success in tasks requiring grounding or reasoning. However, there is currently no systematic framework to organize these agents and plan for future developments. In this paper, the authors propose Cognitive Architectures for Language Agents (CoALA), drawing on the rich history of cognitive science and symbolic artificial intelligence. CoALA describes a language agent with modular memory components, a structured action space for interacting with internal memory and external environments, and a generalized decision-making process for choosing actions. By using CoALA, the authors retrospectively survey and organize a large body of recent work in this field. They also prospectively identify actionable directions for developing more capable language agents. The proposed CoALA framework contextualizes today's language agents within the broader history of AI and outlines a path towards achieving language-based general intelligence. This comprehensive framework provides a foundation for organizing existing research efforts and guiding future developments in the field of language agents.