AI AGENTS

rllm

by rLLM

agentlead-sourcegithub-agent-framework-repositoriessource:github.comreinforcement-learningllmevaluationtraininggithub

Overview

rLLM is a project for training AI agents with reinforcement learning for LLM-based workflows.

Details

rLLM describes itself as “Democratizing Reinforcement Learning for LLMs.” Its official documentation says it lets users train AI agents with reinforcement learning, evaluate and train through a CLI, and use built-in benchmarks plus catalog-resolved agents and evaluators. The docs also define AgentFlow and Evaluator as the two protocols used to run agents on tasks and score the resulting episodes.

When to Use

Use rLLM when you want to train AI agents with reinforcement learning in an LLM-oriented workflow. Use it when you need CLI-based evaluation or training with built-in benchmarks. Use it when you want a framework organized around agent task execution and episode scoring protocols.

Getting Started

Open the official documentation at https://docs.rllm-project.com/.
Review the AgentFlow and Evaluator core concepts to understand how agents run on tasks and how episodes are scored.
Use the documented CLI flow to evaluate or train an agent.
Inspect the GitHub repository at https://github.com/rllm-org/rllm for source code and project updates.

Key Features

•Reinforcement-learning training for AI agents
•CLI-based evaluation and training
•Built-in benchmarks
•Catalog-resolved agents and evaluators
•AgentFlow protocol for running agents on tasks
•Evaluator protocol for scoring resulting episodes

Capabilities

•reinforcement-learning
•agent-training
•agent-evaluation
•cli-workflows
•benchmarks

Last updated Jun 2, 2026