Google Colab GPU Status
Monitor your Google Colab GPU/TPU/CPU runtimes, live NVIDIA VRAM usage, and Compute Units (CU) directly from the status bar in VS Code, Google Antigravity IDE, Cursor, Windsurf, and VSCodium — with one-click runtime provisioning, browser launch, and stop controls.
100% native TypeScript extension: no Python and no external colab CLI required.
Key Features
- Live Hardware & VRAM Telemetry — see active hardware (
T4, L4, A100, H100, G4, TPU v5e-1, TPU v6e-1, CPU), machine shape (Standard / High-RAM), and live GPU memory usage (used / total GB and %) queried natively via the Colab runtime Jupyter proxy.
- Busy-Kernel Awareness — detects when a notebook/training kernel is
busy without blocking or polluting your execution history (silent: true).
- Compute Units (CU) Balance & Burn Rate — displays remaining Colab Compute Units and hourly burn rate (
CU/hr) in the status bar tooltip so you never accidentally leave an expensive GPU running.
- One-Click Runtime Controls — hover over or click the status bar item to start a GPU/TPU/CPU runtime (including High-RAM presets), open the active VM in your browser, refresh immediately, or stop one/all active sessions.
- Dual Native Authentication — sign in via built-in Google OAuth 2.0 (PKCE) stored in VS Code
SecretStorage, or automatically pick up existing gcloud Application Default Credentials (ADC).
- Works Locally & on Remote Clusters — supports automatic localhost callback on desktop editors and seamless URL/code paste fallback on Remote SSH, Coder, and container workspaces.
Quick Start
- Install the extension from VS Code Marketplace or Open VSX.
- Sign In by clicking
Colab: Sign In in the bottom-left status bar (or running Colab Status: Sign In with Google from the Command Palette):
- Google OAuth 2.0: approve access in your browser. On a remote cluster/SSH session, copy the redirected
http://127.0.0.1:.../callback?code=... URL (or code) from your browser address bar and paste it into the input prompt already open in your editor.
- Or
gcloud ADC (zero-config): if you already use Google Cloud CLI, authenticate once in your terminal and the extension detects it automatically in auto mode:
gcloud auth application-default login \
--scopes=openid,https://www.googleapis.com/auth/cloud-platform,https://www.googleapis.com/auth/userinfo.email,https://www.googleapis.com/auth/colaboratory
- Monitor & Control:
- Hover over the status bar indicator to view full GPU model, VRAM, Compute Units balance, and one-click action links.
- Click the status bar when disconnected to open the QuickPick runtime launcher (GPU / TPU / CPU + High-RAM).
- Click Open in Browser to attach the Google Colab web UI directly to your active runtime, or Stop VM to terminate it.
Commands
Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and search for Colab Status:
| Command |
Description |
Colab Status: Start Runtime... |
Pick and provision a GPU, TPU, or CPU runtime (Standard or High-RAM). |
Colab Status: Refresh Status |
Immediately poll active runtimes, VRAM, and Compute Units. |
Colab Status: Open Active Session in Browser |
Open the active Colab runtime directly in your browser. |
Colab Status: Stop Session |
Stop the active runtime (or choose which runtime / stop all when multiple are running). |
Colab Status: Sign In with Google |
Sign in via Google OAuth 2.0 (stored encrypted in SecretStorage). |
Colab Status: Sign Out |
Remove the stored Google OAuth session from SecretStorage. |
Colab Status: Open Settings |
Open colabStatus extension settings. |
Settings
| Setting |
Default |
Description |
colabStatus.authMode |
"auto" |
"auto" (use stored OAuth session if signed in, else fall back to gcloud ADC), "oauth2", or "adc". |
colabStatus.pollIntervalSeconds |
15 |
Status polling interval in seconds (minimum 5). |
colabStatus.queryVram |
true |
Query live GPU VRAM via nvidia-smi over the runtime Jupyter WebSocket. |
colabStatus.showComputeUnits |
true |
Show Compute Units balance and hourly burn rate in the tooltip. |
colabStatus.oauthClientId |
"" |
(Optional) Override with your own Google OAuth 2.0 Desktop Client ID. |
colabStatus.oauthClientSecret |
"" |
(Optional) Client Secret for a custom Google OAuth Desktop Client. |
colabStatus.oauthConfigPath |
"" |
(Optional) Absolute path (or ~/...) to a custom Google OAuth client JSON file. |
colabStatus.oauthScopes |
["openid", "...userinfo.email"] |
OAuth 2.0 scopes requested during Google Sign In. |
FAQ & Troubleshooting
- Why is there no "Sign Out" link when using
gcloud ADC?
gcloud credentials live in your system's ~/.config/gcloud/application_default_credentials.json and may be shared with other CLI tools, so the extension intentionally never deletes or revokes them. Use Switch to OAuth in the tooltip to sign in with a separate OAuth session, or run gcloud auth application-default revoke in your terminal to revoke ADC.
- What does
Colab A100 | Busy mean?
Your Colab kernel is currently executing code (e.g. a training loop). The extension skips nvidia-smi until the kernel becomes idle so it does not queue behind your job.
- How does Sign In work on Remote SSH / Coder / containers?
Your local browser handles the Google sign-in screen. If
127.0.0.1 port forwarding is unavailable, simply copy the redirected http://127.0.0.1:.../callback?code=... URL from your browser address bar and paste it into the input box shown in the editor.
Contributing & Releases
npm install
npm run typecheck
npm test
npm run build
npm run package
Releases are automated via .github/workflows/release.yml on v* tags (or manual dispatch):
COLAB_OAUTH_CLIENT_ID / COLAB_OAUTH_CLIENT_SECRET — optional build-time Desktop OAuth client defaults injected by esbuild so Marketplace/Open VSX builds offer one-click Google sign-in out of the box.
VSCE_PAT — publishes .vsix to the Visual Studio Marketplace.
OVSX_PAT — publishes .vsix to Open VSX Registry (Antigravity, Cursor, Windsurf, VSCodium).
| |