I built TokenBar, a $5 lifetime Mac menu bar app that shows live AI token usage

Nice.

Seeing usage helps — the tricky part is that behavior can still become less consistent over time even when everything looks normal.

This is exactly the kind of tiny utility AI folks actually use. The menu bar is the right surface because token spend is something you want to notice before the invoice lands.

If you keep pushing this, I'd focus on one killer next step: a threshold alert when spend spikes on a model or project.

Nice — this is a genuinely clear “pain → micro-tool” build. Token cost visibility is one of those things people only care about after they’ve burned money once.

Curious, are you planning to expand this into alerts (like spend thresholds / anomaly spikes), or keeping it strictly as a passive visibility layer in the menu bar?

Also I’m running a small experiment with builders working on infra + devtools like this — $19 entry, winner gets a Tokyo trip (flights + hotel). Round 01 is live (100 cap).

Tight scope — menu-bar-first is the right form factor for any observability tool you want glance-able without context switching. I've built similar small-surface utilities (a lightweight memo app that lives one keyboard shortcut away) and the hardest part is usually keeping update frequency accurate without hammering the source API.

Curious on two technical points:

1. How are you sourcing live token counts for Claude / Cursor — log tailing, local API polling, or IDE extension hooks? Claude Code in particular doesn't cleanly expose usage mid-session, so I'm curious what surface you're reading from.

2. Any per-model cost breakdown, or just aggregate tokens? For me the real surprise isn't total usage, it's discovering a single expensive model silently doing 70% of the bill.

That “guessing the bill” pain is real.

Curious — are people using it just for visibility, or actually changing behavior based on what they see?