Token Monitor is a desktop widget that tracks token usage, cost, and account limits across more than 40 AI coding tools and providers, including Claude Code, Codex, OpenCode, Hermes, OpenClaw, Cursor, Cline, and GitHub Copilot. It reads each tool's local session logs or databases through tokscale, updates within seconds of each turn, and breaks usage down by tool, device, model, session, project, or account.
Support differs per tool across three signals: token usage, account limits (session, daily, weekly, billing, and credit windows), and per-session detail. Some sources, such as OpenRouter, Ollama Cloud, Volcengine, Alibaba Cloud, and compatible third-party API gateways, report only limits or balances. Per-session detail shows tokens per prompt and each reply's token split and tools used, read on demand from local transcripts and never synced.
The widget is local-first and needs no server on a single machine. For several machines, each device watches its own logs and sends summary updates to a hub, which pushes aggregated stats to every connected widget over Server-Sent Events. The hub can run inside one widget, as a self-hosted Node process, or as a Cloudflare Worker that also feeds iOS widgets, and a macOS-only iCloud Drive mode syncs without a hub. A headless agent reports usage from servers, SSH hosts, and WSL machines without the desktop app. Prompts, responses, source code, and file contents stay on the machine.
An optional archive keeps observed daily usage after a source tool prunes its own sessions, so heatmaps and trends survive deletion. Usage exports as CSV and JSON, and the widget also surfaces as a macOS menu bar or Windows tray popover, a floating bubble, a screen-edge dock, and native macOS widgets.
Features
- Live token tracking: token counts from 35+ AI tools update within seconds of each turn, with an optional live tokens-per-second readout
- AI tool limits: session, daily, weekly, billing, and credit windows for Claude Code, Codex, Cursor, OpenRouter, GLM, Kimi, and other providers
- Multiple accounts: several accounts per provider, each with its own limits, and one-click switching of the active Codex account
- Cost and cache stats: cost in USD, TWD, HKD, or CNY alongside cache hit and miss splits per tool and model
- Per-session detail: tokens per prompt and per reply, with the tools each reply used, read from local transcripts
- Multi-device sync: a hub inside the widget, a self-hosted Node hub, a Cloudflare Worker, or iCloud Drive
- Headless agent: reports usage from servers, SSH hosts, and WSL without the desktop widget
- Usage history: an activity heatmap, trend charts, and a dashboard with streaks and per-tool and per-model history
- Deleted session archive: keeps daily usage after a source tool prunes its own transcripts
- Data export: writes usage as tool-agnostic CSV and JSON, manually or on a schedule
- Subscription records: compares each account's plan price with the month's usage cost
- Desktop surfaces: menu bar and tray popovers, a floating bubble, an edge dock, native macOS widgets, and opt-in Discord Rich Presence
