The most interesting feature in a macOS menu bar utility is not the volume mixer or the window snapper. It is the panel that watches your AI coding agents. Vorssaint, a free and open-source toolkit published on GitHub, ships a Dynamic Island module that follows Claude, Codex, OpenCode, and GitHub Copilot: plan limits and when they reset, tokens consumed, API value, active models, projects, and live work, plus a notice when a long task finishes.
That is a strange thing to find in a menu bar app. It is also a precise readout of where developer tooling has gone. The utilities Mac users used to buy one at a time (app switcher, clipboard history, window management) now sit behind a single icon, and one of the tabs is a meter for how much AI inference you have burned. Vorssaint treats agent quota the way older tools treated battery percentage. Something you glance at constantly, because running out has consequences.
The meter is the story
Read the feature list closely. Vorssaint reports plan limits and reset windows “where available,” token counts, and an “API value” estimate. For Codex, it surfaces banked resets and lets you use them from the panel, and each extra Codex account (keyed by CODEX_HOME) gets its own limits card. For Copilot, it reads local CLI and app session logs, and its token totals “update when shutdown metrics arrive.” Aggregate API-value estimates, the README says, “use base rates.”
That last clause is doing quiet work. An API-value estimate built on base rates is a model, not a bill. Vorssaint is telling you what your subscription usage would have cost at list prices. For a developer on a flat monthly plan, that number is the only visible signal of marginal cost. The agent does not send an invoice per task. The meter reconstructs one.
This is the economics of AI coding in miniature. Flat-rate plans hide consumption; usage caps reintroduce scarcity at the edges. Tools that make the cap legible become load-bearing. Vorssaint is not the first to try. It is one of the few to fold it into the same panel as fan speed and CPU load, which is the right instinct. Token spend and thermal throttling are both resource problems now.
What Vorssaint actually is
Strip out the AI panel and Vorssaint is a well-scoped competitor to a stack of paid Mac apps. Per-app volume without an audio driver. A system monitor with CPU, GPU, memory, temperature, battery health, and power history. Window snapping, Dock previews, clipboard history, text snippets, a file shelf, an uninstaller, a Homebrew manager, screen capture with offline text recognition, and a fan control beta. The pitch is “one menu bar icon doing the job of a dozen paid Mac apps,” with no account, no telemetry, and no subscription.
The install model is the genuinely novel part. You pick features individually or start from a preset. Uninstalled features stop loading and disappear from the interface; reinstalling restores their settings. Setup asks only for the permissions your choices need. That is a real engineering decision, not marketing. A menu bar app that loads every module at launch is a menu bar app that eats memory and battery. Letting users unload what they do not run is the correct trade.
Privacy claims are specific enough to check. The README says the network is touched only by update checks, the speed test, Homebrew actions, optional lyric lookup, temporary screenshot and recording links, and feedback you send. Permissions are optional, explained in plain words, and the app tells you when a granted permission is no longer needed by anything. Vorssaint is GPL-licensed, with the name, icon, and look covered separately under TRADEMARKS.md, so forks need their own identity. Builds are signed with an Apple Developer ID and notarized.
One caveat the README states plainly and deserves repeating: the microphone mute feature is not a privacy lock. “Siri and some apps may still hear you,” it says. The indicator shows the feature is on, not that every app receives silence. That is an honest disclosure, and it is the kind of thing a paid utility would bury.
Why the AI tab matters for builders
The Vorssaint panel is a small artifact of a larger shift. Agentic coding tools have moved from novelty to daily driver, and the tooling around them has not caught up. Developers now juggle multiple agents across multiple accounts, each with its own quota, its own reset schedule, and its own notion of what a token costs. The README’s mention of separate CODEX_HOME accounts and per-account limits cards is a tell. People are running more than one Codex identity, and they want to see all of them at once.
An API-value estimate built on base rates is a model, not a bill. The meter reconstructs a cost the subscription hides.
That fragmentation is the opening. If Vorssaint’s agent panel works, it points at a gap the labs have not filled: a single, honest view of what your AI subscriptions are actually consuming. Anthropic, OpenAI, and GitHub each expose limits somewhere, but not in one place, and not with a running dollar estimate. A third-party menu bar app filling that gap is a modest event. It is also a signal that agent usage has become routine enough to warrant a dashboard.
The risks are real. Vorssaint reads local session logs for Copilot and relies on “where available” data for plan limits, which means the panel is only as accurate as the logs and APIs underneath it. Base-rate estimates will drift as pricing changes. And a utility that watches your agent activity is, by definition, watching your work. The local-first design is the right answer, but it puts the burden of trust on the source, which is why the GPL license and the notarized builds matter more here than they would in a clipboard manager.
What to watch
The interesting question is not whether Vorssaint ships a polished menu bar app. It is whether the agent panel becomes a category. If developers start expecting a live meter for Claude and Codex the way they expect a battery indicator, the labs will either build it themselves or leave the space to open-source tools. Vorssaint’s README says the project went from first commit to the front of GitHub trending in three days and top of the Swift charts. That is a fast start for a utility. Whether it holds depends on whether the AI tab is a gimmick or the reason people install it.
For now, the meter exists, it is free, and it is honest about being an estimate. That is more than most agent dashboards can say.