Getting started

Four steps, all in the app at /app or with the command line.

1. An organization

Sign in and create an organization, or pick one with the switcher in the top bar. Everything in Artifacts belongs to one organization: its connections, locations, artifacts and API keys. Every member sees all of it.

2. A fallback location

An organization takes writes once it has a fallback: the location every write goes to when nothing nearer matches. Any S3 endpoint works, Tigris or R2 for example; so does a bucket the app creates in a connected cloud.

• Locations → Add an S3 endpoint, tick fallback. See S3 endpoints.
• Or Connections → Connect AWS, GCP or Azure, run the generated script, paste back what it prints, then Locations → Create a bucket. See Connecting clouds.

Add a location in each region your GPUs run in, and writes from there stay in the region. See Locations and placement.

3. An API key

API keys → Create. The secret is shown once. Boxes and scripts send it as a Bearer token; the crate and SexpGPU read it from O41_ARTIFACTS_API_KEY.

export O41_ARTIFACTS_API_KEY=ak_…

4. Point a run at o41://

SexpGPU takes o41:// from its release with o41-checkpoint, which is in review. Until then, use the o41-checkpoint crate, the CLI or the HTTP API.

SEXPGPU_CHECKPOINT_DIR=o41://experiments/run-1234 \
  sexpgpu run train.sx --resume latest

The first launch finds nothing and starts fresh; every relaunch after a preemption continues from the latest checkpoint, wherever it lands. See Checkpointing a run.

Or with plain HTTP, from anything that speaks it:

curl -H "Authorization: Bearer $O41_ARTIFACTS_API_KEY" \
  "https://artifacts.041.io/api/v1/artifacts/list?under=experiments/"