All pages

Docs · IndexMarkdown

Documentation

rrun ("remote run") runs a folder's command on the cheapest machine that fits, from one GPU for a quick experiment to eight for a big run, on any supported cloud you connect. Logs stream back while it runs, and the machine is deleted when nothing needs it.

The o41-rrun command line is the main way in, and it is built for agents as much as for people. The web app at /app does the same things. Training runs there are a good fit for SexpGPU; their curves go to Metrics by 041 and their checkpoints to Artifacts by 041.

rrun is an early release. For a bug or anything missing, such as a cloud or a GPU, run o41-rrun feedback "<what you need>" --reply-to <you@company.com> or use the web app's Feedback page. For help, email niko@041.io.

Start

Getting started

Install the CLI, sign in, connect a cloud, run a folder and follow its logs.

A job from start to finish, and what happens when something goes wrong.

Use

What gets packed, the command, env and secrets, statuses and failure reasons, restarts, logs and telemetry.

Machines and costs

Offers and prices, how a machine is picked, spot, idle machines and --keep, reserved machines of your own, the sweeper, disks and limits.

API keys and agents

Driving rrun from an agent or CI: API keys, --json, exit codes and environment variables.

The pages at /app and the CLI command each one matches.

Setup

Connect Vast, RunPod, Verda, Google Cloud and Nebius, one or more accounts each; turn accounts on and off for GPU and CPU jobs; Google Cloud and Nebius keyless, through rrun's OIDC issuer.

Keep job logs and telemetry in your own S3-compatible bucket.

Reference

Every o41-rrun command and flag, output and exit codes.

The routes under /api/v1, authentication and errors.