AgentsMeetRL
A curated, code-analyzed index of open-source repositories that train LLM agents using reinforcement learning, organized by taxonomy such as base frameworks, search and RAG, web and GUI, tool use, and multi-agent RL.
AgentsMeetRL is an awesome list that summarizes open-source repositories for training LLM agents using reinforcement learning. A project qualifies if it involves multi-turn interactions or tool use, including tool-integrated reasoning projects, and entries are identified through code analysis of the source repositories with manual review, though the authors note some omissions may remain.
It organizes projects by a taxonomy covering base RL training frameworks such as veRL, OpenRLHF and trl, general and multi-task agents, search and RAG agents, web and GUI agents, tool-use agents, code and software-engineering agents, reasoning agents, multi-agent RL, memory agents, embodied agents, domain-specific agents, reward and training methodology, safety-focused RL, and vision-language-model agents, and links technical details such as the RL algorithms, rewards and environments each project depends on.
It is updated regularly and accepts new project submissions through issues or pull requests.