# Gen-Verse/OpenClaw-RL

OpenClaw-RL: Train any agent simply by talking

Repository: https://github.com/Gen-Verse/OpenClaw-RL
Canonical: https://ross.abutalabs.com/products/openclaw-rl
Homepage: https://arxiv.org/abs/2603.10165
Language: Python
License: Apache-2.0
License Family: permissive
Topics: async, memory-systems, open-claw, openclaw-skills, rlhf, sglang, skill-learning, slime, on-policy-distillation, grpo, coding, gui-application, tinker
Last push: 2026-05-23T04:17:31+00:00

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 83, release rhythm 35, longevity 13
- inputs: {"age_days": 188, "days_push": 102, "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 5655, forks 609 (observed 2026-08-28T04:09:27.169772+00:00)

## What it is
OpenClaw-RL is a framework for training personalized AI agents through reinforcement learning using natural conversation as feedback. It uses a server-client architecture to extract evaluative and directive training signals from live agent interactions, enabling agents to improve simply by being used.

## Use cases
- train a personal AI agent by talking to it
- run agentic RL for terminal, GUI, SWE, and tool-call tasks
- fine-tune LLMs with RLHF from real user interactions
- do on-policy distillation of agent policies
- optimize an agent from group feedback from multiple people
- set up async RL training with sglang or slime

## When to choose
- you want an agent that improves from real-world usage without curated datasets
- you need scalable agentic RL across terminal, GUI, and coding environments
- you want hybrid evaluative and directive RL signals in one training loop
- you prefer self-hosted RL training with zero API dependency

## When to avoid
- you need simple supervised fine-tuning without RL infrastructure
- you lack GPU resources or a serving backend for policy inference
- you want a plug-and-play chatbot rather than a training framework
- your agent stack is incompatible with OpenClaw or the supported RL backends

## Facets
- artifact type: framework
- maturity: active
- function: llm-training, agent-framework, machine-learning, rag, mcp
- domain: reinforcement-learning, large-language-models, machine-learning, developer-tools
- platform: python, cloud
- tags: rlhf, grpo, on-policy-distillation, sglang, slime, tinker, agentic-rl, skill-learning, async-training, personalized-agents, openclaw, ai-agents, linux, gpu, docker

## Member repositories
- Gen-Verse/OpenClaw-RL (main) score 52

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:27.169772+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 2026-08-29T17:55:14.993915+00:00, confidence not recorded.
  - readme: https://github.com/Gen-Verse/OpenClaw-RL (fetched 2026-08-28T04:09:27.169772+00:00, sha cdcfa3ea6351)
  - homepage: https://arxiv.org/abs/2603.10165 (fetched 2026-08-29T08:49:41.724720+00:00, sha c5bf0cede341)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T08:49:41.734205+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T08:49:41.738159+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T08:49:41.740104+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T08:49:41.736261+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.