Copilot Live Cost & Token Tracker

Know exactly what GitHub Copilot is costing you — every token, every prompt, every tool call — live in your status bar as you work.

Copilot bills by tokens and credits, but VS Code never shows you where it goes. This extension does: a real-time gauge in your status bar, a multi-tab dashboard that breaks spend down by model, by prompt, and by token type, and budget alerts before you blow past your limit. Every figure is computed 100% locally — no account login, no telemetry, no data ever leaves your machine.

Why you'll want it

• 🔬 Total token transparency — fresh input, cache read, cache write, and output tokens are tracked and priced separately, never lumped into one opaque number.
• 🔍 Drill all the way down — from a session, to a single prompt, to the model calls and tool calls inside it.
• 💸 No more surprise bills — watch session, daily, and weekly spend update live while you chat.
• 🚦 Stop before you overspend — color-coded warnings and notifications at thresholds you set.
• 🔒 Truly private — reads only local Copilot data; nothing is uploaded, ever.

Features

🔬 Transparent token & pricing breakdown

Copilot's cost isn't one number — it's different token types billed at very different rates, and this extension keeps them separate:

• Fresh input, cache read, cache write, and output tokens are each tracked and priced independently — so you can see how much of your bill is cache-write overhead versus real output.
• All rates are USD per 1M tokens, visible and overridable. Nothing is hidden behind a black box, and models we can't price are clearly badged rather than silently dropped.

🔍 Drill down from session to tool call

• Session view — recent chat sessions, each with its own running cost and turn count.
• Per-prompt breakdown — expand a session to see the cost of every individual user prompt.
• Inside a prompt — open any prompt to inspect its model calls, tool calls (arguments, results, errors), and subagent activity, each with its own token and cost contribution.
• One click on the status bar or a session row jumps straight to the relevant detail.

📊 Multi-tab dashboard

A live dashboard in the Activity Bar sidebar (plus a full-window panel), refreshed automatically as you work:

• Activity tab — token cards for fresh input / cache read / cache write / output, plus a cost-per-prompt table grouped by session.
• Cost tab — KPI cards (Today, This Week, selected range, context weight) and a cost-history line chart.
• Models tab — a doughnut chart and table breaking spend down per model.
• A 7 / 30 / 90-day range selector drives every card, chart, and table.

🧮 Status bar gauge

• Always-visible readout of session cost, today's cost, and current context weight.
• Hover for the full picture: session / daily / weekly spend, model turn counts, today's token split, and optional conversion to your local currency.
• The background turns yellow at your warning threshold and red at your limit.

🚦 Budget alerts

• Warning and limit thresholds for session, daily, and weekly spending.
• A VS Code notification fires the moment a threshold is crossed — once per threshold, re-arming only after spend drops back below it.

🕑 Session history that survives resets

• Automatically detects your current chat session and summarizes recent ones.
• View the last 7, 30, or 90 days in the dashboard, with daily cost buckets.
• Daily aggregates are saved to disk, so your history survives Copilot database resets — with configurable retention (7–365 days, default 90).

🎯 Accurate model pricing

• Built-in pricing for current Claude (Opus, Sonnet, Haiku), the GPT-5 family, Gemini, and more.
• Fuzzy name matching handles version suffixes and brand-new releases automatically, falling back to the closest known sibling. Anything inferred or unknown is badged (see Reading the dashboard).
• Add your own rates for custom or unlisted models via pricing overrides.

🔒 Private & portable

• Works across VS Code, VS Code Insiders, VSCodium, and portable installs on macOS, Windows, and Linux — the storage location is resolved automatically, with nothing to configure.

Requirements

• VS Code 1.125 or later — including Insiders, VSCodium, and portable installs
• GitHub Copilot Chat extension installed (this is where the usage data comes from)
• Node.js 24 on your PATH — the extension reads agent-traces.db through a background worker that runs your host's system Node.js, not the runtime bundled with VS Code. Node 24 (the version the native SQLite binding is built for) must be installed and on PATH.
• OpenTelemetry tracing enabled in Copilot Chat (github.copilot.chat.otel.dbSpanExporter.enabled) — this is what populates the data source. Run Copilot Live Cost & Token Tracker: Enable OpenTelemetry Tracing, or accept the prompt the extension shows on first run.

Installation

From the Marketplace

1. Open VS Code
2. Go to the Extensions view (Ctrl+Shift+X / Cmd+Shift+X)
3. Search for "Copilot Live Cost & Token Tracker"
4. Click Install

From VSIX

1. Download the .vsix file from Releases
2. In VS Code, open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P)
3. Run Extensions: Install from VSIX…
4. Select the downloaded file

How to Use

1. Install the extension — it activates automatically when VS Code starts
2. Open the sidebar — click the Copilot Cost icon in the Activity Bar (left side)
3. Use Copilot as normal — the dashboard updates every 10 seconds (configurable)
4. Monitor the status bar — session and daily cost are always visible at the bottom

Reading the dashboard

Switch between the Activity, Cost, and Models tabs, and use the 7 / 30 / 90-day selector to set the range for every card, chart, and table.

To drill down, expand a session to see its individual prompts, then open a prompt to inspect its model calls, tool calls (arguments, results, errors), and subagent activity — each with its own token and cost contribution.

Per-model rows may carry a small badge:

• estimated — the model wasn't in the built-in pricing table, so its rate was inferred from a related model family. The cost is approximate.
• unpriced — no pricing could be resolved at all, so the model's tokens are shown but contribute $0 to totals. Add a custom pricing override to price it.

Commands

Configuration

All settings are under copilotLiveCostTracker.* in VS Code Settings:

Numeric settings are clamped to safe ranges, so an out-of-range value falls back to a sensible default rather than breaking polling.

Custom pricing overrides

Use pricingOverrides to price new or custom models. Keys are model-name patterns; rates are USD per 1 million tokens. Entries that don't match this shape are ignored.

"copilotLiveCostTracker.pricingOverrides": {
  "my-custom-model": {
    "input": 3.0,      // required: fresh input tokens
    "output": 15.0,    // required: output tokens
    "cached": 0.3,     // required: cache-read tokens
    "cacheWrite": 3.75 // optional: cache-write tokens
  }
}

Data Sources

The extension reads from local Copilot databases in VS Code's user storage. The location is resolved automatically per product flavour and OS (no path is hardcoded):

Known Limitations

• Cache-write token data is only available from agent-traces.db (not debug logs)
• Pricing for very new or custom models may require manual overrides
• Cost figures are estimates based on public API pricing — actual billing may differ

Troubleshooting

The dashboard says no data is available. Cost tracking reads agent-traces.db, which Copilot Chat only writes when its OpenTelemetry tracing is enabled. Run Copilot Live Cost & Token Tracker: Enable OpenTelemetry Tracing (or enable github.copilot.chat.otel.dbSpanExporter.enabled yourself), run a Copilot Chat session, and restart VS Code if the file still isn't created. Until then, you can set copilotLiveCostTracker.costDataSource to with-fallback to estimate from debug logs (cache-write data will be missing).

A model shows as unpriced or estimated. This is expected for brand-new or custom models. Add a pricing override to price it accurately.

Nothing updates / errors in the logs. The extension reads the SQLite databases through a background worker that uses your system Node.js. Check the "Copilot Live Cost & Token Tracker" output channel for details, then use Copilot Live Cost & Token Tracker: Refresh Cost Data to retry.

Contributing

Contributions are welcome! See CONTRIBUTING.md for the full development workflow, build and test scripts, and guidelines.

License

MIT © erviAI