Artifacts

Every run gets a .openresearch/artifacts directory. Any files your run writes there are automatically persisted to cloud storage. Useful for model checkpoints, training logs, plots, or anything you want to keep beyond the run.

Usage

Write files to .openresearch/artifacts/ from your run command. No setup needed.

#!/bin/bash
set -e

python train.py --epochs 10 --checkpoint-dir .openresearch/artifacts/checkpoints
python eval.py --output EVAL.md
cp training.log .openresearch/artifacts/

Files sync in the background every 10 seconds while the run is active, so artifacts are preserved even if the run fails partway through.

Viewing artifacts

Open any run and click the Artifacts tab. Use the refresh button to check for new files while a run is still going.

Downloading artifacts

Use rclone to download artifacts from your terminal. Tokens are long-lived and inherit your project access.

1. Install rclone

brew install rclone   # macOS
# or see rclone.org/downloads

2. Create an access token

Go to Settings → Access Tokens and create a token. Copy it (you'll only see it once) and the rclone config command shown next to it.

3. Configure rclone

rclone config create openresearch webdav \
  url=https://api.openresearch.sh/webdav/ \
  vendor=other \
  bearer_token=YOUR_TOKEN

4. Download

# List your projects
rclone lsd openresearch:

# Download all artifacts from one run
rclone copy openresearch:<projectId>/runs/<runId>/artifacts/ ./out/

# Download all artifacts from every run in a project
rclone copy openresearch:<projectId>/runs/ ./out/

# Mirror, deleting local files no longer on the server
rclone sync openresearch:<projectId>/runs/ ./out/

rclone resumes interrupted transfers, parallelizes downloads, and verifies sizes automatically.