Toketer 发布 VS Code 扩展，在编辑器内实时显示 AI 提示词成本

Toketer — your AI fuel gauge ⛽

See how many tokens each AI prompt used and what it cost — live, right inside your editor.

Toketer is a tiny status-bar meter that shows your AI coding spend at a glance. No website, no login, no setup. Install it and the meter appears immediately.

• 🟢 Zero config — works the instant you install it.
• 🔒 Private by design — metadata only. Toketer never reads or sends your code, files, or prompts.
• 🧮 Offline-first — costs are computed locally from a bundled pricing table.
• 🌍 Works in VS Code and Cursor — no native modules, only public editor APIs.
• 🆓 Free, local, and offline — no account or backend needed.

💬 Real-time cost in the chat panel: just type @toketer

Most developers live in the right-side chat panel and want the cost of every prompt, live. Toketer does this with a chat participant: type @toketer before your prompt (it's "sticky", so you type it once and stay on Toketer).

@toketer explain what a closure is

Toketer forwards your prompt to the model you picked in the panel's dropdown, streams the answer straight back into the panel, and updates the meter with the real model + real token counts — every prompt, in real time.

The honest caveat: Toketer can only meter turns sent to @toketer. It cannot read the default Copilot turn (no public API exposes another extension's AI usage to third parties — this is a platform limit no extension can get around). So chat with @toketer when you want live cost tracking. On hosts without the chat API (some Cursor/VSCodium builds) this feature is skipped and the other sources still work.

What it looks like

• Status bar: ⛽ Toketer: $0.15 today (or ₹12.40 today). Click it to open the panel.
• Cost panel: this session's cost, today's cost, a per-model bar chart, and your most recent prompts with individual token counts and costs — all in plain language.

When your editor doesn't report exact tokens, Toketer shows a clearly-labelled estimate (a ~ prefix and an est badge). Toketer never shows a fabricated precise number.

The honest bit: where numbers come from (and their limits)

Automatic sources — where Toketer gets your usage

Toketer captures AI usage automatically from the places developers actually spend tokens, and feeds it all into the same meter. It's honest about precision: some sources are exact, others are necessarily estimates or manual imports.

Why the honesty? Third-party extensions cannot silently observe another extension's AI usage (e.g. Copilot Chat's or Cursor's native AI) through any public API. So Toketer captures what it can read locally and clearly labels the rest. Open Toketer: Sources any time to see each source's live status.

See PRIVACY.md for the data guarantee and CONTRIBUTING.md for the architecture.

Two audiences, served honestly: Claude Code (exact live) vs. GitHub Copilot

Toketer covers both without faking precision it doesn't have:

• Claude Code users get EXACT cost, automatically and live from local session logs.
• GitHub Copilot users get a LIVE ESTIMATE, plus EXACT import. Copilot runs inside Copilot and never exposes its token usage to any third-party extension — a hard VS Code platform limit. So Toketer does two honest things:
  • Live estimate (on by default): it detects each agent turn from the file edits the agent makes (reading only the size of each change, never the content), counts it as one Copilot premium request (Copilot bills per request, not per token), and — right after the turn settles — announces ~cost · ~token range (guess) in the status bar and adds a ~ copilot (estimated) row to the panel. Tokens are shown as a range labelled "guess" (e.g. ~1k–3k), never a fake precise number, because the agent's hidden context is invisible to us. Think of it as a gauge, not a bill.
  • Exact import: run Toketer: Import Copilot Usage with the report from GitHub → Settings → Billing → Copilot → Usage for the precise figures. Toketer remembers your last import and reminds you (at most weekly) when stale.
• Want exact live per-prompt cost? Type @toketer in chat — those are Toketer's own requests, so it meters the real tokens.

Why can't the estimate appear inside Copilot's reply? Because that reply is rendered by Copilot, not Toketer — no extension can write into another participant's response. That's why the estimate shows in the status bar + panel.

Where to get the Copilot usage report

GitHub → Settings → Billing & licensing → Copilot → Usage → Download usage report. Save the CSV (or JSON), then run Toketer: Import Copilot Usage and pick the file. Re-importing is safe — already-seen rows are skipped.

Feeding Toketer activity

Any tool can record a metadata-only event. From another extension:

await vscode.commands.executeCommand('toketer.recordAiActivity', {
  modelId: 'gpt-4o',
  promptTokens: 1200,
  completionTokens: 800,
  isEstimate: false, // omit or set true if these are estimates
});

Or via a deep link from a CLI/agent (all parameters are metadata only):

vscode://toketer.toketer/activity?model=gpt-4o&prompt=1200&completion=800&estimate=0

Commands

All commands are available from the Command Palette under Toketer:

Does Toketer work in Cursor and other editors?

Toketer is a standard VS Code extension that uses no native Node modules, so it installs and runs in VS Code, Cursor, VSCodium, and other VS Code-based editors. The meter, panel, and commands work everywhere.

However, no editor — VS Code or Cursor — exposes its native AI chat's token usage to third-party extensions. So Toketer cannot silently meter the Copilot Chat panel or Cursor's built-in AI. It meters activity that is reported to it: your own Toketer: Ask AI requests, or events pushed by a companion tool/agent via the command or URI handshake above. Non-VS-Code tools cannot load a .vsix at all.

Settings

Development

npm install
npm run watch     # bundle with esbuild in watch mode
# then press F5 in VS Code to launch the Extension Development Host
npm test          # run the unit tests (vitest)
npm run typecheck # strict TypeScript check
npm run lint      # ESLint
npm run vsce:package # produce a .vsix

See CONTRIBUTING.md for a plain-English tour of the code and a list of good first issues.

License

MIT. Toketer is open source — contributions welcome!