Gemini 3.8 Token Monitor & Automatic Handoff

A lightweight context tracker, compression monitor, and automatic handoff extension for Google Antigravity in VS Code.

Overview

Gemini 3.8 compresses context at roughly ~170,000 tokens (and subsequent cycles at ~340k, ~510k, ~680k).

This extension enables you to perform an orderly handoff before compression occurs, letting you transition into a clean context with an observable handoff artifact.

(Note: Future versions will support additional models).

How It Works

Tokens are estimated within ~2% using a calibrated 3.8 characters per token formula tailored for code, JSON tool schemas, and indentation.

• Sub-millisecond execution: Runs in-memory in pure JavaScript with zero lag.
• Zero dependencies: No external daemons, background services, or compilers required.

How to Use It

1. Live Status Bar Widget

Look at the bottom tray in VS Code:

• Displays active tokens and target: AGY: 89k handoff 150k
• Color coding: Shifts from Green (safe context under handoff target) → Yellow (at handoff target e.g. 150k when handoff directive is injected) → Red (past 170k compression cliff).
• Hover tooltip: Shows breakdowns for User, Model, and Thinking tokens, plus exact headroom remaining.

2. Status Bar Click Menu

Click the status bar item to open quick actions:

• Inject handoff message next turn: Commands the agent to summarize and hand off on its immediate next response.
• Edit handoff message template: Customize the macro-level handoff prompt injected to the assistant.
• View detailed token report: Opens an output panel with full session analytics.
• Change target threshold: Select presets (140k, 150k, 160k, 320k) or enter any custom token number.

Configuration Settings

Available under antigravity.tokenMonitor in Settings (Ctrl+,):

License

MIT License © 2026 Jimmy Jordan