# OpenManus/OpenManus-RL

A live stream development of RL tunning for LLM agents

Repository: https://github.com/OpenManus/OpenManus-RL
Canonical: https://ross.abutalabs.com/products/openmanus-rl
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-05-05T06:03:26+00:00

## Health v2 (maintenance only)
Score: 46/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 80, release rhythm 8, longevity 38
- inputs: {"age_days": 545, "days_push": 120, "days_rel": 482, "gap_med": null, "n_releases_24m": 1}
- flags: prerelease_only
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4155, forks 592 (observed 2026-08-28T04:08:36.553737+00:00)

## What it is
OpenManus-RL is an open-source project for reinforcement-learning-based tuning of LLM agents, built on the verl training framework. It provides agent trajectory datasets, training pipelines, and benchmark evaluation on agent tasks like GAIA, AgentBench, WebShop, and OSWorld.

## Use cases
- fine-tune an LLM agent with reinforcement learning
- train agents to reason and use tools with RL
- evaluate tuned models on agent benchmarks like GAIA and OSWorld
- get an open SFT dataset of agent trajectories
- experiment with rollout strategies and reward models for agents

## When to choose
- you want to RL-tune an LLM agent rather than just prompt it
- you need an open agent trajectory dataset for training
- you want to reproduce or extend reasoning-model-style RL training for agents

## When to avoid
- you only need to run an agent without any training
- you lack GPU resources for RL fine-tuning
- you need a production-ready, stable framework rather than a live research project

## Facets
- artifact type: library
- maturity: active
- function: llm-training, reinforcement-learning, agent-framework, machine-learning
- domain: large-language-models, reinforcement-learning, machine-learning
- platform: python
- tags: rlhf, agent-tuning, verl, post-training, llm-agents, benchmarks, sft-dataset, ai-agents, gpu, linux

## Member repositories
- OpenManus/OpenManus-RL (main) score 46

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:36.553737+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 2026-08-29T18:22:52.000478+00:00, confidence not recorded.
  - readme: https://github.com/OpenManus/OpenManus-RL (fetched 2026-08-28T04:08:36.553737+00:00, sha 8545dcf1ba65)
- Data as of 2026-08-30T08:39:29.467469+00:00.