Tally is a quiet meter for Claude. Real-time usage, per-project budgets, model-aware messages-left, and a portable Snapshot library that moves your context to ChatGPT, Gemini, or Grok in one click.
Watching tokens at
Watch tokens accrue the moment a request resolves. Sub-second latency from Claude completion to the inline bar under your message box.
Opus burns ~5× faster than Haiku. Tally weighs every estimate by the model you're on, so the number always matches your actual headroom.
Set a soft cap per project. Quiet warning at 80%, hard pause at 100% if you want one. CSV export for finance.
Send a Claude chat — including every HTML or code file Claude built — to ChatGPT, Gemini, or Grok. Tally auto-pastes into the input on arrival.
A countdown that actually counts down. Native menu-bar app dims when you're rate-limited so you know when to come back.
Counts run in your browser. Nothing leaves your machine until you opt into sync. Snapshot library lives in chrome.storage.
A single hairline beneath your input. Opus burns ~5× faster than Haiku and the model tag is right there, so the number always matches what you're actually doing.
We hit the 5-hour limit at 4:47 every afternoon and never knew why. Tally showed us the cache was busting every time we touched a file. One config change, three more hours of headroom a day.
Sara Köhler — Staff Eng, Halftone
Live meter, model-aware msgs-left, 2 cross-LLM handoffs every day. Unlimited copy + download.
Unlimited cross-LLM handoffs to ChatGPT, Gemini, Grok. Or leave a review for 3 days of Pro, free.
Local-first. Read-only by default. We never see your prompts — only the counts.
Start counting, free