Developers juggling multiple AI coding subscriptions now have a new way to monitor their usage limits. Pacer, a lightweight macOS menu bar utility, displays two rings that track session and weekly consumption for tools like Claude Code, Codex, and Cursor. The tool projects whether your current burn rate will exhaust your quota before the reset window closes, offering immediate visual feedback on subscription health.

Projecting Burn Rates, Not Just Usage

Unlike standard usage meters that show static percentages, Pacer focuses on velocity. The rings fill based on the projected percentage at reset if the current pace continues. Green indicates a safe margin under 80%, amber warns of approaching limits up to 100%, and a solid red disc signals that the user is on pace to run out exactly at reset. This distinction is critical because 16% usage might seem low until a user realizes they are only 10% into the billing week. The tool provides actionable advice derived from actual model mix data. It suggests specific adjustments, such as reducing usage of high-cost models like Fable or shifting Opus workloads to Sonnet. For Claude users, Pacer calculates per-model weights by joining official usage percentages with local transcripts, fitting weights from data since Anthropic does not publish how different models count against limits. When data is insufficient, it defaults to API price ratios.

Multi-Provider Support and Installation

Pacer aggregates data from every logged-in account on the machine, displaying them side by side. It supports Claude Code, Codex, Antigravity, Cursor, Gemini CLI, and GitHub Copilot. The application is built with minimal dependencies, consisting of one Node script and one Swift file. Installation is straightforward via Homebrew with brew install dkremsa/tap/pacer, followed by starting the service to ensure it survives reboots. Users can also build from source, which compiles the app into /Applications and installs a launchd agent. The tool stores state in a JSON file at ~/.claude/pacer/status.json, refreshed every 10 minutes. This file includes pace, level, advice, and cost metrics, allowing other tools to read the data. It tracks API-equivalent costs over rolling day, week, and 30-day periods, helping developers understand the real value of their subscriptions. Notifications trigger when the pace crosses 100% or when any window passes 90%, ensuring users are alerted before hitting hard limits.

Key Takeaways

  • Pacer projects usage velocity rather than just displaying current percentages.
  • The tool supports multiple AI coding providers including Claude, Codex, and Cursor.
  • Installation is simple via Homebrew, with a lightweight architecture using Node and Swift.
  • Per-model weights for Claude are calculated from local transcripts and official usage data.
  • Notifications alert users when burn rates threaten to exceed subscription limits before reset.

The Bottom Line

For developers managing tight AI coding budgets, Pacer transforms opaque usage limits into actionable, real-time data. It’s a practical tool for preventing unexpected throttling during critical development cycles.