Kaggle: Sync & Run

Develop locally, run on Kaggle compute. Push notebooks/scripts, choose CPU/GPU/TPU, attach datasets, and auto-download outputs — all from VS Code.

Requirements

• Kaggle CLI installed: https://github.com/Kaggle/kaggle-api (ensure kaggle is on your PATH)
• A Kaggle API token

Authentication

• Preferred: set env var KAGGLE_TOKEN_JSON to your kaggle.json content (e.g., { "username": "...", "key": "..." }).
• Or use “Kaggle: Sign In” (prompts for username and key). Token is stored in VS Code Secret Storage.
• “Kaggle: Sign Out” clears the stored token.

Quick start

1. Sign In: open any Kaggle view (Runs / My Notebooks / Datasets) and use the three-dots menu → Kaggle: Sign In.
2. Init: run “Kaggle: Init Project” to create kaggle.yml and kernel-metadata.json.
3. Open a .ipynb and click the rocket “Run on Kaggle”.
4. Outputs auto-download to .kaggle-outputs (configurable). You can also run “Kaggle: Download Outputs”.

Features

• Notebook toolbar button: “Run on Kaggle” (respects kaggle.yml accelerator/internet; prompts only if missing)
• Auto-poll and download outputs when run completes
• My Notebooks view: list your kernels; click to download/open locally under remote_notebooks/<user__slug>
• Datasets view: list datasets; right-click to Attach, Browse Files (preview/download a single file), or Download all
• Runs view: shows recent run URLs (from .kaggle-run.log)

Key commands

• Kaggle: Sign In / Sign Out
• Kaggle: Init Project
• Kaggle: Link to Existing Notebook
• Run on Kaggle (toolbar/menus)
• Kaggle: Download Outputs
• Kaggle: Attach Dataset / Attach from tree
• Kaggle: Browse Dataset Files / Download Dataset
• Kaggle: Submit to Competition

Configuration

• kaggle.cliPath: path to Kaggle CLI (default: kaggle)
• kaggle.defaultAccelerator: none | gpu | tpu (default: none)
• kaggle.defaultInternet: boolean (default: false)
• kaggle.outputsFolder: outputs download folder (default: .kaggle-outputs)
• kaggle.autoDownloadOnComplete: auto download outputs (default: true)
• kaggle.pollIntervalSeconds / kaggle.pollTimeoutSeconds: output polling behavior

Project files

• kaggle.yml — extension config (code_file, accelerator, internet, privacy, datasets, competitions, outputs)
• kernel-metadata.json — Kaggle notebook metadata read by the CLI

Example kaggle.yml

project: my-project
kernel_slug: yourname/my-awesome-kernel
code_file: notebook.ipynb
accelerator: gpu
internet: false
privacy: private
datasets: []
competitions: []
outputs:
  download_to: .kaggle-outputs/

Folders used by the extension

• .kaggle-outputs/ — run outputs (auto-downloaded)
• .kaggle-datasets/<user__dataset>/ — dataset downloads
• remote_notebooks/<user__slug>/ — notebooks pulled from My Notebooks

Troubleshooting

• “Not found” in My Notebooks: we fallback to kaggle kernels list --mine. Ensure you’re signed in.
• CLI missing: set kaggle.cliPath to the full path of the CLI.
• No outputs: ensure the run finished or increase timeouts (kaggle.pollTimeoutSeconds).
• Auth: you can set KAGGLE_TOKEN_JSON or use Sign In (username/key prompt). Sign Out clears stored token.

Development & testing

• Build: npm run compile
• Watch: npm run watch
• Tests: npm test (integration tests via @vscode/test-electron; optional E2E require CLI + creds)

Security notes

• API token is kept in VS Code Secret Storage. The extension never writes kaggle.json to disk.