AgentR1/Agent-R1

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1633 stars · 113 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Agent-R1 is a modular Python framework for training LLM agents with end-to-end reinforcement learning. It models each interaction turn as a step-level MDP transition, making tool use, environment feedback, context management, and reward assignment explicit parts of the training loop.

Use cases

• train an LLM agent to use tools via reinforcement learning
• run multi-step agentic RL with environment feedback
• fine-tune models on HotpotQA, ALFWorld, or WebShop agent tasks
• apply StepPO-style step-level policy optimization
• perform online policy distillation for agents
• build custom agentic RL training environments

When to choose

• you need step-native RL training for multi-turn LLM agents rather than single-turn pipelines
• you want explicit control over tool use, context management, and reward assignment during training
• you want a research framework with recipes for standard agent benchmarks

When to avoid

• you only need single-turn RLHF or DPO fine-tuning without agent interaction
• you need a production inference or agent-serving framework rather than a training framework
• you lack GPU resources for RL training

Facets

framework · maturity active

agent-framework llm-training reinforcement-learning machine-learning artificial-intelligence reinforcement-learning large-language-models machine-learning python agentic-rl step-level-mdp tool-use multi-step-agents policy-optimization online-policy-distillation ai-agents gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="AgentR1/Agent-R1")