OhMyToken

Real-time token dashboard for AI coding assistants -- a KDE Plasma 6 widget that tracks usage, quotas, throughput, costs, and sessions across multiple AI tools, right from your desktop panel.

Supported Tools

Tool	Data Source	Real-time	Quotas
Claude Code	~/.claude/ telemetry, sessions, JSONL	/proc/pid/io 1s polling	Session window + daily (auto-detected tier)
Gemini CLI	~/.gemini/ chat sessions	/proc/pid/io child worker polling	Requests/day by tier
Pi	~/.pi/agent/sessions/ JSONL	/proc/pid/io polling	Token + cost tracking
OpenCode	~/.local/share/opencode/ SQLite DB	/proc/pid/io polling	Token tracking
Antigravity IDE	Language server local API	Status endpoint polling	Prompt + Flow credits
Gemini API	countTokens endpoint (free)	--	Rate limits (requests + tokens)

Each tool can be individually enabled or disabled. The widget gracefully handles tools that aren't installed or aren't running.

Features

Live Tachometer

Car-style animated gauge with rotating needle, engine vibration jitter, three color zones (green/yellow/red), and adaptive max scale. Responds in real-time to streaming activity via /proc/pid/io polling at 1-second intervals.

Dual Quota Rings

Concentric input/output progress rings with color-coded thresholds -- green under 70%, yellow 70-90%, red above 90%. Session window quotas on the left, daily totals on the right.

Dashboard Layout

Session Ring | Tachometer | Daily Ring side by side, with session countdown timer below. Falls back to a standalone tachometer when quota data isn't available.

Hourly & Daily Charts

• Hourly: 12-hour smooth Catmull-Rom curves with adaptive bucket aggregation (5/10/15/20/30/60/120 min)
• Daily: 8-day stacked bar chart (input + output)

Active Sessions

Live session cards with token counts, duration, message count, and pulsing glow animation that reflects streaming activity per-session.

Model Breakdown

Per-model token usage with proportional bars and cost estimates.

Summary Tab

Aggregated view across all enabled providers -- total tokens today/week/month, combined sessions, merged throughput rates.

Panel Indicator

Two modes for the system tray:

• Quota Ring -- mini session progress ring with percentage label
• Mini Tachometer -- animated gauge showing real-time throughput across all active tools

Cost Tracking

Estimated API costs (weekly and total) with configurable monthly budget threshold.

Installation

From .plasmoid file

bash build.sh
kpackagetool6 -t Plasma/Applet -i ohmytoken.plasmoid

From GitHub Releases

Download the latest ohmytoken.plasmoid from Releases, then:

kpackagetool6 -t Plasma/Applet -i ohmytoken.plasmoid

Upgrade

bash build.sh
kpackagetool6 -t Plasma/Applet -u ohmytoken.plasmoid

Development (symlink)

ln -s /path/to/plasmoid-ohmytoken/ohmytoken ~/.local/share/plasma/plasmoids/ohmytoken

Changes to source files are live immediately -- just reload the widget.

Configuration

Right-click the widget and select Configure....

Services

Setting	Default	Description
Claude Code	On	Monitor ~/.claude/ telemetry and sessions
Gemini CLI	On	Monitor ~/.gemini/ chat sessions
Pi	On	Monitor ~/.pi/agent/ sessions and usage
OpenCode	On	Monitor ~/.local/share/opencode/ SQLite database
Antigravity	On	Monitor via language server API (auto-discovered)
Gemini API	Off	Requires API key -- get one at ai.google.dev

Display

Setting	Default	Description
Refresh interval	300s	Data polling interval (60--900s)
Panel indicator	Ring	Quota ring or mini tachometer in panel mode
Show costs	On	Display estimated API costs
Monthly budget	$100	Budget threshold for cost tracking

Claude Limits

Setting	Default	Description
Daily input limit	Auto	Override auto-detected tier limit (0 = auto)
Daily output limit	Auto	Override auto-detected tier limit (0 = auto)

Tier limits are auto-detected from ~/.claude/.credentials.json:

Tier	Input/day	Output/day
Max 5x	1.665B	166.5M
Max	300M	30M
Team	200M	20M
Pro	100M	10M

Gemini CLI tier is detected from ~/.gemini/settings.json (Free: 1000, Standard: 1500, Enterprise: 2000 requests/day).

Architecture

local_stats.py        ----> JSON ----> main.qml (Claude)
gemini_local_stats.py ----> JSON ----> main.qml (Gemini CLI)
pi_stats.py           ----> JSON ----> main.qml (Pi)
opencode_stats.py     ----> JSON ----> main.qml (OpenCode)
antigravity_stats.py  ----> JSON ----> main.qml (Antigravity)
gemini_stats.py       ----> JSON ----> main.qml (Gemini API)
/proc/pid/io          ----> grep ----> main.qml (tachometers)

• Backend: Python scripts parse local data files, SQLite databases, and local APIs. Output JSON to stdout.
• Real-time: /proc/pid/io rchar polling at 1s for Claude, Gemini CLI, Pi, and OpenCode processes. Language server endpoint polling for Antigravity.
• Rendering: Split Canvas layers (static background / dynamic arcs) with GPU-composited needle rotation for minimal CPU impact. Jitter animation only runs during active streaming.

Requirements

• KDE Plasma 6
• Python 3
• Linux (requires /proc filesystem for real-time I/O monitoring)
• One or more of: Claude Code, Gemini CLI, Pi, OpenCode, Antigravity IDE

License

GPL-3.0+

Credits

Built with Claude Code