# rllm-org/rllm

Democratizing Reinforcement Learning for LLMs

Repository: https://github.com/rllm-org/rllm
Canonical: https://ross.abutalabs.com/products/rllm
Homepage: https://docs.rllm-project.com
Language: Python
License: Apache-2.0
License Family: permissive
Topics: agent-framework, agentic-workflow, distributed-training, llm-training, machine-learning, ml-infrastructure, ml-platform, reinforcement-learning, coding-agent, llm-reasoning, search-agent, swe-agent, tinker, verl
Last push: 2026-08-24T06:46:08+00:00

## Health v2 (maintenance only)
Score: 77/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 69, longevity 41
- inputs: {"age_days": 584, "days_push": 9, "days_rel": 125, "gap_med": 56, "n_releases_24m": 4}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5803, forks 611 (observed 2026-08-28T04:09:29.744048+00:00)

## What it is
rLLM is an open-source Python framework for training language agents with reinforcement learning, supporting any agent harness (Claude Code, LangGraph, OpenAI Agents SDK, custom agents) and multiple training backends (verl, tinker, fireworks) switchable with one flag. It includes a CLI for evaluating models on 60+ integrated benchmarks and wraps agent code with @rllm.rollout to trace LLM calls for RL training.

## Use cases
- train an LLM agent with reinforcement learning like GRPO or RLOO
- evaluate a model on SWE-bench or Terminal-Bench from the CLI
- fine-tune a coding agent on software engineering tasks with RL
- run RL training on math benchmarks like AIME and MATH-500
- wrap an existing LangGraph or OpenAI Agents SDK agent for RL training
- run distributed multi-GPU RL training with verl or single-machine training with tinker
- benchmark LLM agents across 60+ tasks with auto-pulled datasets

## When to choose
- you want to RL-train agents without rewriting them for a specific harness
- you need to switch between distributed multi-GPU and single-machine training backends easily
- you want integrated evaluation across many agentic, math, and coding benchmarks
- you need sandboxed rollouts at training scale with Docker, Modal, or Daytona

## When to avoid
- you only need simple supervised fine-tuning without agentic rollouts
- you have no GPU or API budget for large-scale rollout generation
- you need a lightweight inference-only serving framework rather than a training framework

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-training, reinforcement-learning, benchmarking, cli, sdk
- domain: reinforcement-learning, large-language-models, machine-learning, developer-tools
- platform: python
- tags: rl-training, agentic-rl, verl, tinker, rollout, swe-agent, grpo, llm-evaluation, sandboxed-execution, ai-agents, docker, gpu, linux, macos

## Member repositories
- rllm-org/rllm (main) score 77

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:29.744048+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:52:45.484078+00:00, confidence not recorded.
  - readme: https://github.com/rllm-org/rllm (fetched 2026-08-28T04:09:29.744048+00:00, sha 051f3130fdc4)
  - homepage: https://docs.rllm-project.com (fetched 2026-08-29T08:47:53.425375+00:00, sha 9fde7160523e)
  - site_page: https://docs.rllm-project.com/installation (fetched 2026-08-29T08:47:53.435553+00:00, sha 9a05a787e727)
  - site_page: https://docs.rllm-project.com/quickstart-cli (fetched 2026-08-29T08:47:53.438336+00:00, sha 4350d2a41b68)
- Data as of 2026-08-30T08:39:29.467469+00:00.