A new open-source hardware project called Clawdmeter is giving Claude Code power users a physical way to track one of the most invisible costs of AI-assisted development: token usage.
Built around a small ESP32-powered AMOLED display, Clawdmeter sits on a desk and shows real-time usage statistics from Claude Code, including session activity, weekly utilization, and token-related signals. Instead of keeping usage dashboards buried inside a browser tab or terminal window, the project turns AI consumption into something developers can see at a glance.
Created by Reykjavik-based developer Hermann Haraldsson, the device has quickly attracted attention from the AI developer community, gaining hundreds of GitHub stars and multiple forks within days of launch. Its appeal is simple: as coding agents become part of everyday software work, developers increasingly want better visibility into how much AI they are using, how quickly they are burning through limits, and when a session is getting expensive or intense.
A DIY Gadget for the Claude Code Era
Clawdmeter is not a commercial device in the traditional sense. It is an open-source firmware project designed for off-the-shelf hardware, most notably a Waveshare ESP32-S3 board with a 2.16-inch AMOLED screen.
Users buy the hardware themselves, flash the firmware, configure it with their Claude Code credentials, and pair it with their laptop over Bluetooth. Once set up, the device becomes a tiny desktop companion for monitoring Claude Code usage in real time.
The project fits neatly into the growing culture around AI-assisted coding, where developers are no longer only asking whether a model can write or debug code. They are also tracking how efficiently it works, how many tokens it consumes, and how much of their available usage window remains.
That behavior has become common enough to earn its own informal vocabulary, including “tokenmaxxing,” a term used by AI-heavy developers who obsessively monitor and optimize their token consumption.
How Clawdmeter Works
Technically, Clawdmeter combines a small embedded display, Bluetooth connectivity, and Claude Code usage data into one compact interface.
The device connects to a laptop using Bluetooth Low Energy and uses the ESP32-S3’s capabilities to communicate with the user’s development environment. Under the hood, it relies on tools such as LVGL for graphics and NimBLE for Bluetooth functionality. It can also emulate a Bluetooth keyboard, allowing its physical buttons to send shortcuts back to the connected machine.
The device reads a user’s Claude Code OAuth token and calls the Claude Code API to fetch usage data. It then extracts relevant metrics from API response headers and renders them directly on the AMOLED display.
That setup allows Clawdmeter to stay synchronized with actual Claude Code limits and session behavior rather than simply estimating usage locally.
Pixel Art Meets Token Tracking
One of the reasons Clawdmeter has spread quickly is that it does not present itself as a dry developer meter.
On startup, the device displays small pixel-art animations featuring a character called “Clawd.” The animation changes depending on usage intensity, becoming more energetic as token consumption rises. The result is a playful visual cue that makes usage feel more immediate without forcing the developer to read a dense dashboard.
The central hardware button lets users cycle through different animation modes and usage views. Beyond the animated display, Clawdmeter can show session-level statistics, weekly utilization charts, Bluetooth status, and reset options.
This mix of utility and personality is part of the charm. It feels closer to a desk toy for serious AI users than another analytics screen.
Hardware Buttons Double as Claude Code Shortcuts
Clawdmeter also includes physical controls that interact with Claude Code through Bluetooth keyboard emulation.
The side buttons can send keyboard shortcuts to the paired computer. By default, they trigger actions such as Space or Shift+Tab combinations, which can be mapped to Claude Code workflows like activating voice mode or toggling interaction modes.
That turns the device into more than a passive display. It becomes a small hardware controller for AI-assisted coding sessions.
For developers who already keep Claude Code running throughout the day, this kind of physical shortcut layer adds a surprisingly practical benefit. It reduces the friction of switching modes, starting voice interactions, or controlling the coding agent without constantly returning to the keyboard.
Why Developers Are Paying Attention
Clawdmeter arrives at a moment when AI coding tools are becoming more continuous and agentic.
Developers are no longer using AI only for occasional code snippets. Tools like Claude Code increasingly sit inside full development workflows, helping with debugging, refactoring, test generation, file edits, and project-wide reasoning. As these sessions get longer, usage tracking becomes more important.
A normal chatbot conversation may not make token visibility feel urgent. A multi-hour coding session with an agent working across a repo does.
That is where Clawdmeter’s physical format makes sense. It provides a persistent ambient signal. A developer does not need to open a usage page, inspect logs, or wait for an error message. The device quietly shows how active the session is and how much usage is being consumed.
Open Source Makes It Easy to Modify
Because Clawdmeter is open source, its long-term value may come less from the first version and more from what developers build on top of it.
The firmware can be forked, customized, and extended. Developers can add new animations, build additional usage screens, adapt the device for other AI platforms, or connect it to different APIs. A similar concept could easily be applied to OpenAI, Gemini, local LLM runtimes, or multi-provider coding environments.
That flexibility is important because the AI tool market is changing quickly. Developers often use multiple assistants and model providers at once. A small physical dashboard that starts with Claude Code could eventually become a broader AI usage meter for the entire desktop workflow.
Security Is the Main Caveat
The project does come with an obvious security consideration.
Clawdmeter needs access to a Claude Code OAuth token to retrieve usage data. That means users must treat the device and its firmware with the same caution they would apply to any tool with account-level API access.
Because the project is open source, technically skilled users can inspect how the firmware handles credentials. But the security burden still matters. Anyone building or modifying the device should understand where tokens are stored, how they are transmitted, and what happens if the hardware is lost or compromised.
For casual users, this may be too much setup. For developers comfortable flashing firmware and managing credentials, the tradeoff will feel more reasonable.
A Small Sign of a Bigger AI Hardware Trend
Clawdmeter is a tiny gadget, but it points to a larger pattern.
As AI tools move deeper into daily work, developers are beginning to want physical interfaces around them. Not every AI interaction belongs in a browser tab. Some information is better displayed ambiently, especially when it involves usage limits, cost, session activity, or agent status.
That is why Clawdmeter feels interesting beyond its novelty. It represents a small step toward physical dashboards for AI work, where models, tokens, agents, and automation become part of the developer’s desk environment.
The project is not trying to replace Claude Code’s interface. It is trying to make the hidden rhythm of AI-assisted coding visible.
For now, Clawdmeter is a niche DIY tool for people deep enough into Claude Code to care about live token tracking. But the idea behind it is likely to age well. As AI coding agents become more autonomous, more expensive, and more central to software work, developers will want better ways to monitor them. Sometimes, that may mean another dashboard. Sometimes, it may mean a tiny glowing screen on the desk showing exactly how hard your AI assistant is working.
Post Comment
Be the first to post comment!
