#4949th of 53 in Training and fine-tuning
openclaw-rl
Asynchronous RL framework that trains agents from live conversation feedback
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 5.7k
- License
- Apache-2.0
- Last commit
- May 2026
- Language
- Python
4,992 on 21 Apr 20265,744 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
OpenClaw-RL wraps a self-hosted model as an OpenAI-compatible API for OpenClaw, intercepts multi-turn conversations, and trains the policy in the background using Binary RL (GRPO), on-policy distillation, or a hybrid of both. Built on the Slime framework, it also includes RL recipes for terminal, GUI, SWE and tool-call agents. Training can run on local GPUs or on Tinker (LoRA).
Who it is for: ML engineers doing RL fine-tuning of agents on their own GPUs
Strengths
- Serving, rollout, judge scoring and training run as separate async loops, so serving continues during training
- Supports Binary RL, on-policy distillation and a combined method, with LoRA and Qwen3.5 support
- Includes launch scripts for terminal, GUI, SWE and tool-call agent RL
- Policy, judge and trainer run on your own infrastructure; no third-party model API required
Weaknesses
- Default setup assumes 8 GPUs, CUDA 12.9 and Python 3.12
- Depends on Slime and Ray-style launch scripts; setup is involved
- Personal-agent flow is tied to OpenClaw and needs an extension for your own install
- No Docker or compose files, and no tagged releases
What it needs
- GPU required
- Needs Slime, OpenClaw, CUDA 12.9, Python 3.12
- Models: Qwen3 4B/8B, Qwen3.5 4B/9B/27B, Qwen3-VL 8B
- port 30000
Quick answers
Is openclaw-rl open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Does openclaw-rl need a GPU?
Yes, it needs a GPU.
Also in Training and fine-tuning
See all 53| Rank | Project | Score |
|---|---|---|
| 1 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 2 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 88 out of 100 |
| 3 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 86 out of 100 |
| #4 | verlReinforcement learning training library for LLM post-training with PPO and GRPO | 86 out of 100 |
| #5 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 86 out of 100 |