Agent Reinforcement Trainer (ART)
Train, evaluate, and iterate on LLM agents in hours, not weeks
Agent Reinforcement Trainer (ART) is an open-source framework for training agentic LLMs to improve performance and reliability through experience. It provides a lightweight interface to reinforcement learning (RL) algorithms such as Group Relative Policy Optimization (GRPO), boosting model quality while minimizing training costs. ART integrates with W&B Training Serverless RL, eliminating provisioning and management overhead and providing instant access to elastic GPU capacity.
RL made accessible to everyone
Reinforcement learning is still early, and the algorithms are intricate. Small, nuanced mistakes can prevent convergence or cause instability. ART abstracts the complexity behind a developer-friendly API, so you can train models without wrestling with fragile scripts or algorithmic details.
Why ART?
Built for real-world agents
Real user interactions are multi-turn. ART supports multi-turn rollouts so your agent learns from realistic conversations and performs reliably.
Drop‑in integration
OpenAI‑compatible chat endpoint slots straight into your existing code or frameworks like CrewAI, OpenAI Agents SDK, and LangGraph.
Works flexibly with existing code
ART provides wrappers to plug RL training into existing apps and abstracts the training server into a modular service your code needn’t touch.
The Weights & Biases end-to-end AI developer platform
Weave
Models
The Weights & Biases platform helps you streamline your workflow from end to end
Models
Experiments
Track and visualize your ML experiments
Sweeps
Optimize your hyperparameters
Registry
Publish and share your ML models and datasets
Automations
Trigger workflows automatically
Weave
Traces
Explore and
debug LLMs
Evaluations
Rigorous evaluations of GenAI applications
