# mll-lab-nu/RAGEN

Agent RL framework for LLM agents: multi-turn reinforcement learning with StarPO and reasoning-collapse diagnostics

Repository: https://github.com/mll-lab-nu/RAGEN
Canonical: https://ross.abutalabs.com/products/ragen
Homepage: https://ragen-ai.github.io/
Language: Python
License: MIT
License Family: permissive
Topics: agent-rl, agentic-ai, llm-agents, reinforcement-learning, rl-framework, multi-turn-rl
Last push: 2026-08-23T22:59:46+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 41
- inputs: {"age_days": 585, "days_push": 10, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2778, forks 227 (observed 2026-08-28T04:07:21.534771+00:00)

## What it is
RAGEN is a Python framework for training reasoning LLM agents with multi-turn reinforcement learning using the StarPO algorithm. It also provides diagnostics for studying failure modes in agentic RL, such as the Echo Trap and reasoning/template collapse.

## Use cases
- train llm agents with multi-turn reinforcement learning
- apply starpo to trajectory-level agent training
- diagnose reasoning collapse in agent rl training
- detect template collapse where outputs ignore inputs
- research failure modes in agentic reinforcement learning
- run ppo or grpo with snr-adaptive prompt filtering
- reproduce ragen paper experiments

## When to choose
- you need to RL-train multi-turn LLM agents with trajectory-level rewards
- you want research-grade diagnostics for agent RL stability
- you want to reproduce or extend the RAGEN/StarPO papers

## When to avoid
- you only need inference-time agent orchestration without RL training
- you need a production-ready, turnkey RLHF pipeline for chat models
- you lack GPU resources for large-scale LLM training

## Facets
- artifact type: framework
- maturity: active
- function: reinforcement-learning, llm-training, agent-framework, machine-learning, benchmarking
- domain: reinforcement-learning, large-language-models, machine-learning
- platform: python
- tags: agent-rl, starpo, multi-turn-rl, reasoning-collapse, llm-agents, rl-training, diagnostics, research-framework, ai-agents, research, gpu, linux, docker

## Member repositories
- mll-lab-nu/RAGEN (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:21.534771+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 2026-08-30T08:16:30.094706+00:00, confidence not recorded.
  - readme: https://github.com/mll-lab-nu/RAGEN (fetched 2026-08-28T04:07:21.534771+00:00, sha c9793a8f5a01)
  - homepage: https://ragen-ai.github.io/ (fetched 2026-08-29T09:55:45.483878+00:00, sha 89fe183c2a49)
- Data as of 2026-08-30T08:39:29.467469+00:00.