https://github.com/ccnets-team/remoterl
RemoteRL — zero-setup cloud service for reinforcement learning, anywhere and at any scale
https://github.com/ccnets-team/remoterl
cloud-native gymnasium reinforcement-learning remote-control rllib stable-baselines3
Last synced: 20 days ago
JSON representation
RemoteRL — zero-setup cloud service for reinforcement learning, anywhere and at any scale
- Host: GitHub
- URL: https://github.com/ccnets-team/remoterl
- Owner: ccnets-team
- License: other
- Created: 2025-06-28T16:17:50.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2025-08-12T07:41:49.000Z (about 2 months ago)
- Last Synced: 2025-09-13T01:37:49.430Z (28 days ago)
- Topics: cloud-native, gymnasium, reinforcement-learning, remote-control, rllib, stable-baselines3
- Language: Python
- Homepage: https://remoterl.com
- Size: 321 KB
- Stars: 3
- Watchers: 0
- Forks: 0
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE.txt
Awesome Lists containing this project
README
# RemoteRL – Remote Reinforcement Learning for Everyone, Everywhere 🚀
> **Cloud-native RL in a single line of code**
[](https://pypi.org/project/remoterl/)
[](#)
[](#)
[](https://pypi.org/project/gymnasium/)
[](https://pypi.org/project/websockets/)
[](#)
[](#)
[](https://pypi.org/project/remoterl/)
* **[Installation](#-installation)** · **[Configure Key](#-configure-your-api-key)** · **[Hello‑World Example](#-hello-world-example)** · **[Next Steps](#-next-steps)**
---
## 🧩 How It Works — Three Pieces Mental Model
Simulator(s)/Robot(s) ⇄ 🌐 RemoteRL Relay ⇄ Trainer (GPU/Laptop)
> The **trainer** sends actions, the **simulator** steps the environment, and the relay moves encrypted messages between them. Nothing else to install, no ports to open.
>
> * **Isolated runtimes** – trainer and simulator can run different Python or OS stacks.
> * **Elastic scale** – fan in 1…N simulators, or fan out distributed learner workers.
> * **Always encrypted, never stored** – payloads travel via TLS and are dropped after delivery.
> * **Free tier:** every account includes **1 GB of data credit(per week)** (≈ 1 M CartPole steps).
---
## 📦 Installation
```bash
# Gymnasium only (lightweight)
pip install remoterl
# + Stable‑Baselines3
pip install remoterl stable-baselines3
# + Ray RLlib
pip install remoterl "ray[rllib]" torch pillow
```
---
## 🔐 Configure Your API Key
To use RemoteRL, you need an API key.
You can get it either via CLI or from the website:
### Option 1 — CLI (Recommended)
```bash
remoterl register # Opens browser and fetches your API key automatically
```
The key will be saved to your local config automatically.
### Option 2 — Manual (for server, CI, or scripts)
1. Visit [remoterl.com/signup](https://remoterl.com/signup) and **sign up for an account**
2. Go to your Dashboard
3. Copy your API key
Set it as an environment variable:
```bash
export REMOTERL_API_KEY=api_xxxxx...
```
## 💻 Hello World Example
### Run **two terminals**:
```bash
# Terminal A – simulator
$ remoterl simulate
# Terminal B – trainer
$ remoterl train
```
remoterl simulate — Python example
```python
import remoterl
# 1. Decide at runtime whether this process is the trainer or the simulator
remoterl.init(role="simulator") # blocks
remoterl.shutdown() # optional
```
remoterl train — Python example
```python
import gymnasium as gym
import remoterl
remoterl.init(role="trainer") # one call switches to remote mode
env = gym.make("CartPole-v1") # actually runs on the simulator
obs, _ = env.reset()
for _ in range(1_000):
action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)
if terminated or truncated:
obs, _ = env.reset()
```
That’s it – you’ve split CartPole across the network.
## ⚡ Latency & Isolation
| Path | Typical RTT | Notes |
| ------------------------------- | ----------- | ----------------------------------- |
| Trainer ↔ same‑region simulator | 10‑50 ms | Feels like local play. |
| Trainer ↔ cross‑continent | 50‑150 ms | Use frame‑skip for twitchy control. |
---
## 📚 Next Steps
* **[`Cloud Service Overview`](<./docs/Overview/overview-cloud-service.md>)** – details on what it is and how it works.
* **[`Console-Output Guide`](<./docs/SDK (Python)/sdk-console-output-guide.md>)** – step-by-step screenshots from a *live* trainer ↔ simulator session, with every line highlighted and explained.
* **[`Quick-Start (Init & Shutdown)`](<./docs/SDK (Python)/sdk-quick-start-init-shutdown.md>)** – step-by-step examples of `remoterl.init()` and `remoterl.shutdown()` for trainers and simulators.
* **[`Trainer Cheat-sheet`](<./docs/SDK (Python)/sdk-trainer-remote-call-cheat-sheet.md>)** – Gymnasium, Stable-Baselines3, and RLlib one-liners for remote execution.
## 📎 Quick Links
- 🔑 [Get your API Key](https://remoterl.com) – Create an account on the official site to get your key.
- 📊 [RemoteRL Dashboard](https://remoterl.com/user/dashboard) – Manage your usage, keys, and settings.
- 📘 [Documentation Index](./docs/Overview/overview-cloud-service.md) – Start from the top-level service overview.
---
## 📄 License
RemoteRL is distributed under a commercial license.
We offer a free tier, while premium plans help offset our worldwide cloud-server costs. See [`LICENSE`](./LICENSE.txt) for details.
---
**Happy remote training!** 🎯