Mfrostbutter

Tokenpulse — AI skill for Claude Code

AI community

Self-hosted AI usage and cost panel for Raspberry Pi.

How to install Tokenpulse

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Mfrostbutter/tokenpulse and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Tokenpulse does

Self-hosted AI usage and cost panel for Raspberry Pi. Tracks spend and quotas across Anthropic, OpenAI, OpenRouter, and Claude Code on a touchscreen, with a web console and Home Assistant integration. Pure-stdlib Python, zero dependencies.

Alternatives in AI

Hindsight — State-of-the-art long-term memory for AI agents by Vectorize 6.7k ★
Manifest — Real-time cost observability for OpenClaw agents — track tokens, costs, messages, and model usage 4.1k ★
WindsurfAPI — Turn Windsurf / Devin Desktop's 100+ AI models (Claude, GPT, Gemini, DeepSeek, Kimi, GLM, SWE) into OpenAI-, A 3k ★

README

TokenPulse

A wall-mounted answer to one question: **what is my AI actually costing me right now?**

TokenPulse is a self-hosted usage and cost panel for AI providers. A pure-stdlib Python daemon polls every provider you give it a key for, a kiosk browser turns a Raspberry Pi touchscreen into an appliance, and the same daemon feeds Home Assistant and any browser on your network. No cloud, no accounts, no third-party services, zero Python dependencies.

Loosely inspired by [vibepulse](https://github.com/Mfrostbutter/vibepulse), an ESP32 panel in the same spirit; TokenPulse is its own build, from scratch.

![Pulse page](docs/panel-pulse.png)

The gauge is whichever limit will hurt first. Here that is a Claude weekly model cap at 95% with the reset time under it, next to today's spend, month to date, and tokens per hour across every source.

Screens

Five pages, swipe or tap to move, auto-rotating by default.


	QUOTA. Every limit from every provider, loudest first. Money-denominated quotas show dollars left, plan limits show percent and reset time. Stale data is flagged, never dressed up as fresh.
	SPEND. Per-provider today and month to date, with a 30-day trend.
	HEAT. Daily peak utilization per quota, GitHub-graph style. Builds as the daemon runs.
	SYS. Source health with data age, host vitals, brightness, manual refresh.

There is also a desktop **management console** at `/admin.html` for everything the touchscreen is wrong for:

![Console](docs/console.png)

What it tracks

Source	What you get	Credential required
Anthropic	Org-wide API spend + token volume, today and MTD	Admin API key (`sk-ant-admin...`)
OpenAI	Org-wide API spend + token volume, today and MTD	Admin API key (`sk-admin-...`)
OpenRouter	Per-key spend, credit balance, and