Detailed Analysis
Throttle Meter is a macOS menu bar application built by a 16-year-old developer to address a specific gap in Anthropic's Claude Code tooling: the absence of a public API or granular interface for monitoring usage limits in real time. The tool reads local session files stored at `~/.claude/projects/*.jsonl` to compute both a 5-hour rolling usage window and a weekly window, surfacing that data directly in the macOS menu bar. Released under the MIT license and written entirely in Swift, it targets macOS 14 and above, requires no signup or network calls, and emits threshold notifications at 80% and 95% consumption. A paid tier (€19, one-time) extends the experience with a Project window, an AI assistant that audits `CLAUDE.md` and related configuration files via actual `read_file` tool calls, and an "Exact mode" that polls live usage data by piggybacking on a Safari claude.ai session — an approach the developer explicitly flags as unofficial and fragile.
The motivation behind Throttle Meter reflects a genuine friction point for Claude Code power users. Anthropic's in-app warning triggers only at 90% usage, and the 5-hour rolling window that governs throttling is not surfaced proactively, meaning developers frequently discover their limits only after they have already been hit. By parsing the JSONL session logs that Claude Code writes locally, Throttle sidesteps the need for any official API surface, making it resilient to Anthropic's silence on the matter. The cost extrapolation feature — projecting usage into EUR equivalents — further suggests the tool targets developers either on paid plans or trying to model what API-equivalent consumption would cost, underscoring how opaque Anthropic's internal rate-limit accounting has been for end users.
The tool's emergence fits squarely within a broader ecosystem of third-party Claude usage monitors that has grown rapidly in 2025 and early 2026. Similar projects include Claude-Code-Usage-Monitor (a terminal-based tool with ML-driven predictions), Dan Butuc's native macOS menu bar app using API interception, and Rich Hickson's Swift-based app that queries Anthropic's usage API via keychain credentials. Commercial products like Usage4Claude and SessionWatcher occupy the same space. This proliferation signals a structural gap: Anthropic has not yet provided developers with a first-party, programmatic view into their Claude Code consumption, creating demand that independent developers are rushing to fill. The variety of approaches — local file parsing, API interception, keychain credential reuse, session cookie hijacking — also illustrates how fragmented and unofficial these solutions remain.
What distinguishes Throttle Meter within this landscape is its strict local-only architecture for the free tier and its transparency about the fragility of its paid "Exact mode." Most competing tools either rely on undocumented API endpoints or require some form of credential access, both of which carry the same brittleness risk the developer openly acknowledges. The decision to keep the core meter functional regardless of whether Anthropic changes internal endpoints is a sensible product architecture choice: the free tier survives on file system data alone, while the more powerful but risky features are isolated to the paid upgrade. The developer's call for community feedback on whether the computed numbers match claude.ai's own display also reflects an honest acknowledgment that without an official API, ground truth verification is itself a community exercise. That a teenager building their first commercial product is navigating these API opacity issues speaks to how broadly felt the tooling gap has become among Claude Code's developer user base.
Read original article →