zjunlp/LLMAgentPapers

Must-read Papers on LLM Agents.

What this project is

LLMAgentPapers is a curated reading list of academic papers about large language model (LLM) agents — AI systems that use a language model as their "brain" to plan, remember, use tools, or talk with other agents. Think of it as a well-organized bibliography for anyone who wants to study this fast-moving research area.

Who made it and when

  • Creator: zjunlp, a natural language processing research group (the name suggests Zhejiang University's NLP lab).
  • Started: June 2023, specifically to track papers about multi-agent systems.
  • Still active: The repository shows a recent commit badge and includes papers dated as late as 2026, so it's being maintained.

What's inside

The list is organized into clear sections:

  1. Overview papers — surveys that map the whole field of LLM agents.
  2. Single-agent papers, broken into five sub-topics:
    • Personality — giving agents traits, roles, or human-like behaviors.
    • Memory — helping agents remember information over long conversations or tasks.
    • Planning — getting agents to break down goals into steps.
    • Tool use — letting agents call external software or APIs.
    • RL training — training agents with reinforcement learning.
  3. Multi-agent papers, split into:
    • Task-oriented communication (agents working together or competing to get things done)
    • Casual/open conversations (agents just chatting)
  4. Applications — real-world uses.
  5. Frameworks — software systems for building agents.
  6. Others — anything that doesn't fit above.
  7. Resources — benchmarks, tool lists, and types of tools.

Each paper entry lists the title, authors, a link to the abstract (usually on arXiv), and the publication date.

Notable recent addition

In March 2024, the team published their own paper, KnowAgent, about using knowledge to improve how LLM agents plan.

Related projects

The README points to two sibling paper lists from the same group:

  • Prompt4ReasoningPapers — papers on reasoning with language model prompting.
  • KnowledgeEditingPapers — papers on editing knowledge inside large language models.

License

The repository is released under the MIT license, meaning it's free to reuse with attribution.

Related

  • Project
    Project-N-E-K-O/N.E.K.O

    N.E.K.O. is an open‑source AI companion that combines voice, text, and visual understanding with a five‑layer memory system and multiple avatar forms. It runs locally (Python 3.11) and can be deployed via one‑click binaries, Docker, or source build. Features include proactive chat, real‑time multimodal interaction, agent‑style tool execution, a plugin marketplace, and support for 14+ LLM providers. The project is Apache‑2.0 licensed and has active Discord/QQ/Steam communities.

  • Project
    Contrastive-LM/CLM

    Contrastive Language Models (CLM) is a genuine AI project offering a fast, contrastively trained decision‑making model (CLM‑8B) that scores actions against textual states. It provides a pip‑installable library, a FastAPI server with a web playground, and a typed‑question API (binary, choice, score). The model achieves state‑of‑the‑art verification performance on agentic benchmarks while being up to 9× faster than comparable agents. It includes scripts for fine‑tuning the lightweight projection heads on custom data and is released under Apache 2.0.

  • Project
    keon/browser-control

    A fast Rust CLI that lets AI coding agents control a real browser via the Chrome DevTools Protocol using shell-native commands and stable element references.

  • Project
    Owl-Listener/designer-skills

    A set of Claude Code/Gemini‑CLI plugins that turn design knowledge into machine‑readable skills and ready‑to‑run commands, letting AI agents execute full design workflows (research, strategy, UI, interaction, systems, ops, critique) directly from the assistant.