ModelBar

ModelBar is a lightweight, provider-extensible macOS menu-bar app for AI model quotas, token usage and service status.

It ships with OpenAI, Claude and Grok adapters. Forks can add any provider that can produce ModelBar's small ProviderReading contract, whether the source is a CLI, HTTP API or local usage file.

What it shows

• quota windows with percentage remaining and reset time
• local token totals for today or the last 24 hours and seven days
• provider service status when a reliable source exists
• optional, read-only Hermes token totals by agent profile
• provider controls, refresh interval and launch-at-login settings

ModelBar uses native AppKit, one status item, no WebViews and no persistent child processes.

Requirements

• macOS 14 or newer
• Swift 6.2 or newer when building from source
• CodexBar CLI for the three built-in provider adapters

The built-in adapters look for codexbar at /opt/homebrew/bin/codexbar or /usr/local/bin/codexbar. CodexBar's Homebrew cask and app settings can install the CLI in those locations.

Hermes is optional. If no live Hermes profile databases exist, ModelBar continues to show provider data.

Build and install

swift test
./Scripts/build-app.sh release
ditto .build/ModelBar.app /Applications/ModelBar.app
open /Applications/ModelBar.app

The local build is ad-hoc signed. It is not notarised or distributed through the Mac App Store.

Add a provider

A provider addition has three parts:

1. implement ProviderAdapter
2. add the adapter to ProviderRegistry.standard
3. add production-shaped fixtures and tests

The menu, settings, cache, refresh filtering, stale-data handling and neutral-brand fallback are provider-independent. A provider may optionally supply separate light and dark accent colours.

See Adding a provider and the copy-ready adapter template.

Privacy and resource use

• ModelBar stores no provider credentials or raw provider responses.
• Provider commands run only during refresh and have explicit deadlines.
• Hermes databases are opened read-only.
• The snapshot cache contains display data only and is written only when content changes.
• The current native build targets effectively zero idle CPU and a physical footprint below 40 MB.

Contributing

See CONTRIBUTING.md. Security reports should follow SECURITY.md.

ModelBar is available under the MIT License.