tokcount — live token & cost for AI chat 作者： vøiddo

tokcount is a live token, cost, and context-window meter for every major AI
chat product. As you type in the composer, a small overlay in the corner
shows:

• tokens — estimated using per-family heuristics (o200k / claude /
gemini / llama / mistral / cohere / gpt) plus language multipliers
for non-latin scripts
• cost — input-token dollar estimate at the current model's per-1M
input rate
• context % — how much of the model's context window you've already
committed to the prompt

It works in:

• ChatGPT (chat.openai.com + chatgpt.com)
• Claude (claude.ai)
• Gemini (gemini.google.com) and AI Studio (aistudio.google.com)
• Perplexity (perplexity.ai)
• Poe (poe.com)
• OpenRouter (openrouter.ai)
• Hugging Face Chat (huggingface.co)
• Grok (grok.com + x.com/i/grok)
• Mistral Le Chat (chat.mistral.ai)
• DeepSeek (chat.deepseek.com)
• Cohere Coral (coral.cohere.com)

WHY USE IT

• Stop writing five paragraphs into ChatGPT then realizing you blew
past the context window — the bar turns red at 85%
• Stop eyeballing "how expensive is this prompt, actually" — it shows
dollars, not abstract token counts
• 30-day usage dashboard by model and project, CSV export (Pro)
• Daily / weekly / monthly budget guards with optional hard-stop (Pro)

FREE FOREVER

• Overlay on every supported platform
• 7-day local history, 200 recent measurements
• Accurate cost math for the 30+ most-used models
• Pricing refreshed from a public manifest every hour — when vendors
cut prices, your extension updates the same day (no reinstall)

PRO — $4.99/mo

• 10-year history, unlimited entries, 50 saved projects
• Budget guards with hard-stop (prompt blocked when over-budget)
• CSV export + cache-hit detector
• Higher-accuracy tokenizer overrides per model family
• Custom model definitions (bring your own model cost row)

Team and Enterprise tiers available at voiddo.com/tools/tokcount/upgrade.

PRIVACY

tokcount does not read your prompts after you send them. It counts
characters in the composer before you hit send, converts locally, and
writes nothing but aggregate daily totals to browser-local storage. The
only outbound requests are:

(1) a once-per-hour pull of the public pricing manifest
(voiddo.com/tools/tokcount/pricing.json — no cookies, no body)
(2) a once-per-day license revalidation IF you've entered a Pro key

No analytics. No telemetry. No third-party scripts. Source at github.com/voiddo.

Full policy: https://voiddo.com/tools/tokcount/privacy/

Made by vøiddo — a six-person studio based in Israel. Questions: support@voiddo.com.