thinkwee/AgentsMeetRL

What it solves

AgentsMeetRL is a curated, open-source collection of projects that use reinforcement learning (RL) to train LLM agents. It addresses the challenge of navigating the rapidly growing landscape of agentic-RL by cataloging repositories that feature multi-turn interactions or tool use, helping researchers and developers understand how different projects implement RL algorithms, reward models, and training environments.

How it works

The project provides a structured taxonomy of over 400 repositories, categorized by their technical focus—such as base training frameworks, search and RAG, web interaction, or self-evolution. Each entry is manually reviewed to ensure it contains actual RL-training code. The repository also functions as a Claude Code skill, allowing users to ask technical questions about agentic-RL implementation, such as debugging reward functions or selecting the right framework, with answers grounded in the documented projects.

Who it’s for

This resource is designed for AI researchers, machine learning engineers, and developers building or studying LLM agents who need to compare technical approaches, find existing benchmarks, or troubleshoot RL training workflows.

Highlights

• Curated Corpus: A comprehensive, manually verified list of 411 open-source projects focused on agentic-RL.
• Technical Taxonomy: Categorizes projects into 16 distinct areas, including Base Frameworks, Reward & Training, Multi-Agent RL, and Embodied agents.
• Claude Code Integration: Includes a dedicated skill (agents-meet-rl) that provides on-demand, expert-level assistance for experiment design and debugging.
• Transparency: Clearly distinguishes between projects with public code and those still under review, ensuring users only find actionable, executable resources.
• Community-Driven: Actively updated with new research and open-source contributions, with a clear process for community submissions.