TokenSight Features Pricing Compare Try it free

Know where your
AI tokens actually go

A Mac menu-bar app for Claude Code and Codex. See every limit, exact token usage, and what's quietly burning your budget — read from the logs already on your Mac, at zero token cost.

Download — 7-day free trial Buy — $9.99 once
brew install --cask sanv725/tap/tokensight
Notarized by AppleNo logins or API keysReads local files onlymacOS 13+, Apple silicon & Intel
TokenSight dashboard: tokens this cycle, Codex and Claude weekly and 5-hour limits, daily usage by tool, and the top Insight TokenSight menu-bar popover with limit ring, Codex and Claude weekly windows and today's cost
Insights

Find what's burning your budget

Totals tell you that you used 94M tokens. Insights tell you why — and what to change.

  • ✓Cache rebuilds — stepping away, switching models or editing CLAUDE.md re-sends your whole context
  • ✓Startup overhead — projects whose memory files and MCP servers cost 40K+ tokens before you type
  • ✓Context hogs — the tool or MCP server behind most of your context growth
  • ✓Long sessions & sub-agents — where a fresh session or a cheaper model pays off
Insight card: 30 cache rebuilds re-sent 2.3M tokens, with the cause, the fix and a Show sessions button
Limits

Every limit, before it bites

Codex's weekly limit and Claude's 5-hour and weekly limits, with reset times — the numbers the servers report — in your menu bar, with alerts at 50, 80 and 95%.

Codex weekly 51%, Claude weekly 44% and Claude 5-hour 38% with reset times
Sessions

Every session, every request

Codex and Claude Code side by side: tokens, models and estimated cost per session, down to a per-request inspector that shows what each call added to context.

Sessions table listing Codex and Claude Code sessions with project, model, requests, tokens and cost

Everything the vendor dashboards won't tell you

Exact totals

Per project, week, model and request — straight from the servers' own accounting, not estimates.

Context attribution

Which built-in tools, MCP servers and shell commands grew your context, with the dominant cause on every request.

Codex + Claude Code

One app for both, each on its own reset clock — including sub-agents and Codex auto-review.

Alerts

Notifications at 50 / 80 / 95% for every limit, and a menu-bar gauge that fills as you go.

Cost estimates

What your usage would cost at API list prices, per session and per day — editable rates.

Your data, your Mac

All-time history in a local database, with CSV and JSON export. Nothing leaves your Mac.

Private by architecture

TokenSight parses local files your tools already wrote. No agent integration, no SDK, no telemetry.

What it reads

  • ~/.codex/sessions — Codex's own session logs
  • ~/.claude/projects — Claude Code's transcripts
  • Your status line data — Claude's limits, only if you turn it on

What it never does

  • Spend tokens — it never calls a model
  • Ask for logins — no passwords, API keys or browser cookies
  • Send your usage anywhere — the only network requests are update checks and a weekly license check

Simple pricing

Every install starts with a 7-day trial of everything. No subscription.

Free

$0
after your trial
  • Current weekly cycle, in full
  • Limits, alerts and menu bar
  • Charts and per-request inspector
  • Top Insight

License

$9.99
one time · perpetual for 1.x
  • Everything in Free
  • All-time history
  • Every Insight, with the fix
  • CSV & JSON export
Buy TokenSight

Up and running in a minute

Install

Download the DMG and drag to Applications — or brew install --cask sanv725/tap/tokensight.

Watch it fill in

It reads the Codex and Claude Code history already on your Mac and keeps counting live. Optionally connect Claude's limits in Settings.

Buy when you're ready

Your 7-day trial includes everything. Afterwards, a one-time license keeps history, export and all Insights.

FAQ

Does TokenSight use my AI plan or tokens?

No. It only reads log files Codex and Claude Code already wrote. It never calls a model.

Do I need to sign in to Claude or OpenAI?

No logins, API keys or browser cookies. Claude's limits come from your Claude Code status line, if you choose to connect it — how that works.

How is this different from ccusage or CodexBar?

Those are great free tools: ccusage for terminal reports, CodexBar for quotas across 40+ providers. TokenSight goes deeper on Codex and Claude Code — which tools fill your context and Insights on what to change. See vs ccusage, vs CodexBar, and all five compared.

What happens after the trial?

TokenSight keeps working for free on the current weekly cycle. A one-time $9.99 license unlocks all-time history, export and every Insight.

Refunds?

Billing, tax and refunds are handled by Polar as merchant of record. If TokenSight doesn't work for you, email us and we'll sort it out.