Live limits, before you're blocked
Your Claude 5-hour and weekly limits with a forecast — “at this pace, ~5 hr to the limit” — a heads-up while there's still room to react, not at 95%.
Open source · macOS menu-bar app
See where your AI tokens went before the bill does.
A local-first command center for AI-assisted development. StackPulse reads the logs Claude Code and Codex already write to disk — no account, no API key, nothing leaves your Mac.
MIT licensedmacOS 14+No telemetryRust + Swift 6
StackPulse
starting…Scanning your Mac…
Reading local logs…
Detecting providers…
Computing usage…
3 of 11 providers detected
One monitor, every provider
Watch Claude, ChatGPT, and Gemini — from one place.
StackPulse reads the logs each tool already writes and turns them into cost, tokens, and live limits — locally, in real time. No API keys, no dashboards to wire up.
Claude
165M tok · full cost scan
ChatGPT · Codex
685M tok · full cost scan
Gemini
sessions · activity
…and detectsCopilotCursorGrokOpenRouterDevinZ.aiOpenCodeGemini CLI
A look inside
It lives in your menu bar. It opens into this.
Home
Today's spend, live limits, active AI sessions, and anything that needs you — the moment you open it.
Diagnostics
Is StackPulse healthy, and if not, exactly what to fix — each issue listed with a repair step and a secret-free support bundle.
Missions
What your agents are doing across Claude, Codex, and Antigravity — observed locally, in real time.
What it shows you
Everything your AI stack is spending — in one place.
Where the money went
Per-model, per-window token and cost for Claude Code and Codex, read from local logs. Every figure is a labeled estimate — never presented as your bill.
What your agents are doing
Live Claude, Codex, and Antigravity sessions observed on this Mac — what each one is working on, right now.
Your projects & stacks
Launch a dev server in one tap, run remembered commands, and see whether each project's stack is current and secure.
Diagnostics & repair
Answers “is StackPulse working, and if not exactly why” — credentials, refresh, permissions, data integrity, each with a fix.
Command Center
A full-screen monitor for a second display — AI usage and mission control at a glance while you work.
Honest by design
A number you can actually trust.
01
Estimates, not your bill
Every dollar figure is computed on your machine against an embedded table of public list prices, always tagged as a local estimate. On a subscription it's an equivalent metered figure — usually far more than you paid.
02
Refuses to guess
If it doesn't recognize a model, it counts the tokens but drops the cost rather than borrowing a sibling's rate — and a “canary” names the gap, so it can say “I might be under-reporting” instead of showing a confidently wrong total.
03
Documents its gaps
It labels every number with where it came from and states its known inaccuracies in the open. The whole design is built around honest reporting.
Eleven providers, honestly scoped
It only claims what it can actually measure.
Full local cost scan
ClaudeCodex
Per-model cost, cache metrics, and limit windows — read from local logs, zero network.
Activity only
Antigravity
Sessions, recency, and model — its tokens live server-side at Google, so cost is not shown.
Detection only
CopilotCursorDevinGeminiGrokOpenCodeOpenRouterZ.ai
Installed / authenticated presence — no usage, no cost, no limits.
Open source · MIT
Every line is on GitHub.
No black box. The portable Rust core makes no network calls at all, and the app is offline by default with a short, named list of opt-in exceptions that go only to the provider — never to us. Read every line, open an issue, or build it yourself.
Read the privacy policy →Stop guessing what your agents cost.
Clone StackPulse, run make run, and watch your AI spend in real time.