# rrun by 041

> Yeet a folder. Run a command. On the cheapest GPU that fits. Say what hardware you need. rrun finds the cheapest machine on the clouds you connect, streams the logs back and deletes the machine when you are done.

Documentation: https://rrun.041.io/docs/index.md. Every page on this site as markdown: https://rrun.041.io/llms.txt. Sign in: https://rrun.041.io/app. Support: niko@041.io. Also by 041: [SexpGPU](https://sx.041.io) for training, [Metrics](https://metrics.041.io) for experiment tracking and [Artifacts](https://artifacts.041.io) for checkpoints.

```console
$ curl -fsSL https://rrun.041.io/install.sh | sh
$ o41-rrun run --gpu RTX5090,RTX5080,A6000 --ram 64 --spot -- python train.py
j-4k2m9x7qpz queued. Follow: o41-rrun logs j-4k2m9x7qpz -f
```

## Why rrun

**Ask for hardware, not for a cloud.**

- **The cheapest machine that fits.** From one RTX 5090 for a quick experiment to eight B300s for a big run. List the GPUs and counts that will do, and rrun takes the lowest price across every cloud you connect, on spot when you want it.
- **Any cloud, one command.** Connect the clouds you have accounts on, and one request searches them all. Yours is not supported yet? Tell us: `o41-rrun feedback "please support <your cloud>" --reply-to you@company.com`
- **Fast while you iterate.** Jobs, status and logs answer in a fraction of a second. A machine stays warm for 5 minutes after a job, or as long as --keep says, so your next job starts in 2–3 seconds.
- **Built for agents.** JSON output, stable exit codes and the job id on stdout. Every page, docs included, comes back as markdown when you curl it, starting at /llms.txt. The web app does exactly what the CLI does.
- **Nothing keeps billing.** When the warm period ends, the machine is deleted with its disks, and rrun sweeps up anything left behind. No forgotten instances.
- **Keys with limits.** Agents and CI get keys that can only manage jobs. Logs stay in built-in storage or your own S3 bucket.

## Start

Bring your cloud accounts. Sign in, connect a cloud, run a folder. Questions: niko@041.io