Tally

See what's left of every AI plan, long before you hit the limit.

Every plan, one bar

Claude, Codex and OpenCode Go sit side by side in your menu bar, each as a small meter of what's used and what's left.

Counted down to the reset

Every session and weekly limit shows when it refills, to the minute, so you know whether to wait or switch.

A warning before the wall

Tally watches your pace and tells you when you'll run dry, while there's still time to finish the task.

What it's really worth

Your usage at API prices: $149 of Claude this week on a $20 plan. Free, open source, macOS 14+.

Click a meter and the panel tells you the rest.

  1. Most headroom

    The plan with the most room right now. Here, OpenCode Go with 98% left.

  2. Running now

    The session in progress and what it has cost so far: $17.55.

  3. Session and weekly

    Percent used, the time to the reset, and a white tick where an even pace would put you.

  4. Runs out in ~4h 11m

    Worked out from your last hour of use, so you hear about it while there's still time to finish.

  5. 19× value

    $88.98 at API prices against $4.67 of a $20 plan, over the same seven days.

The Tally panel: $126.67 spent in 7 days; Claude Pro running now at $17.55 this session, session 13% used with 4h 22m to reset and a warning that it runs out in about 4h 11m, weekly 31%, 19× value; Codex on ChatGPT Plus at 42% weekly, ahead of pace.
Spend by model treemap for 7 days: claude-opus-5-5 70%, $88.97 via Claude; gpt-6.1-sol 13% via Codex; gpt-6.2-sol 8%; smaller shares via OpenAI and OpenCode Go.

Where the money went, by model and by the plan that paid.

A dithered treemap of the week. Here claude-opus-5-5 took 70%, $88.97, all of it on Claude Pro.

Spend also groups by project, folding worktrees into their repo, and by tool, so Codex CLI and Pi on the same plan stay apart.

It reads what your agents already write. Nothing else leaves your Mac.

No account, no analytics, no server. Each login goes only to its own provider, and Tally never refreshes or rewrites one.

Plans it tracks
PlanLive limitsUsage from
ClaudePro, Max5-hour session, weekly, weekly per modelClaude Code
CodexChatGPT Plus, ProSession, weekly, reset creditsCodex CLI, Pi
OpenCode GoRolling 5 hours, weekly, monthlyPi, OpenCode
OpenCode ZenSpend only, free modelsOpenCode
OpenAI APISpend onlyPi
Where each number comes from
DataSource
Spend, tokensSession logs in ~/.claude/projects, ~/.codex/sessions, ~/.pi/agent/sessions and OpenCode's opencode.db
Claude limitsClaude Code's Keychain login, sent to Anthropic's usage endpoint
Codex limits~/.codex/auth.json, sent to ChatGPT's usage endpoint, or Codex's own logs
OpenCode GoThe Go API key from Pi or OpenCode, sent to OpenCode's usage endpoint
PricesThe LiteLLM price table, downloaded once

Running in under a minute.

Paste this into Terminal. It fetches the latest release, checks its checksum, installs Tally into Applications and opens it. The meters appear in the menu bar; there's no Dock icon. The first scan of your logs takes a few seconds.

Terminal
npx tally-menubar
No Node? Use curl instead
curl -fsSL https://github.com/PratikRai0101/tally/releases/latest/download/Tally.zip -o /tmp/Tally.zip \
  && ditto -x -k /tmp/Tally.zip /Applications && open /Applications/Tally.app
  • macOS14 Sonoma or later, Apple Silicon or Intel
  • Node18 or later, for the npx command
  • Signed in toat least one of Claude Code, Codex, Pi or OpenCode
  • PriceFree, MIT licence

Tally isn't notarized by Apple yet. Both Terminal commands work straight away. A zip downloaded in a browser needs one approval: System Settings → Privacy & Security → Open Anyway.