wtf is awesome-llm-agent-papers?

This repository is a curated reading list of more than 200 research papers about LLM agents, meaning language models that are given the ability to plan, remember things, use external tools, and work together with other agents to pursue goals over many steps. It is meant as a companion to a survey paper called "LLM Agents: A Survey," and it is actively kept up to date as new research comes out. The list is organized into ten sections that mirror the survey's structure: general surveys, agent architectures, planning, memory, tool use, multi-agent systems, environments, applications, evaluation, and safety. Each paper entry includes a short note explaining what the paper contributes, along with its publication venue and year, and a link to the official code when one exists. Sections are collapsed by default in the README so a reader can expand only the areas they care about. For someone new to the topic, the project includes a Starter Kit of ten papers chosen as a good entry point, each with a short explanation of why it is worth reading first. Examples include the ReAct paper, which introduced interleaving reasoning with actions, and Generative Agents, which explored memory and reflection for simulated populations of agents. There is also a rotating "papers to watch" section that highlights recent 2026 work drawing early attention, covering areas like token-efficient agents, long-term memory, GUI-controlling agents, and agent safety guardrails. Because this is a plain-text reading list rather than a piece of software, there is nothing to install or run. The value is in the curation itself: contributors add new papers as they are published, each tagged by topic, so someone researching a specific part of agent design, such as memory or safety, can go straight to the relevant section instead of searching the wider literature on their own. The project is licensed under MIT and welcomes pull requests to add new papers, based on a contributing guide in the repository. The full README is longer than what was shown.