glimmer

A pixel-art always-on desk widget. Custom firmware for the GeekMagic SmallTV-Ultra that rotates through glanceable channels — Claude / Codex usage, clock, weather, push cards — with crisp retro typography on a 240×240 panel.

---

What it shows

Channel	What
Home	At-a-glance clock + weather + Claude/Codex usage meters + 24-hour timeline
Clock	Big VT323 digital clock, day-of-week, greeting
Weather Now	Hero temp, feels/humidity/wind, 3-day mini cards
5-day Forecast	Range-bar rows showing min/max + condition
Claude usage	5-hour window % + weekly % + reset countdowns
Codex usage	Primary % + secondary % + credits/reset
AI Today	Combined Claude+Codex card with 7-day mini chart
Info	IP / SSID / signal / uptime / heap / CPU / firmware
Push cards	One-shot notification cards via POST /push

All channels render via region-based partial repaints — no flickering between data updates.

Quick start

git clone git@github.com:<you>/glimmer.git
cd glimmer
pio run -e nodemcuv2 -t buildfs    # build LittleFS image (fonts + web UI)
pio run -e nodemcuv2               # build firmware

# Flash a freshly-stocked SmallTV-Ultra (over your home LAN):
DEVICE_IP=<find via arp or device screen>
curl -F "filesystem=@.pio/build/nodemcuv2/littlefs.bin" http://$DEVICE_IP/update
curl -F "firmware=@.pio/build/nodemcuv2/firmware.bin"   http://$DEVICE_IP/update

# Device reboots into glimmer's setup AP. Connect to "glimmer-setup" Wi-Fi
# (open, no password) and visit http://192.168.4.1/ to enter your home
# Wi-Fi credentials.

For the full flashing dance (including the one-time "get the stock firmware onto your Wi-Fi first" step), see FLASHING.md.

Hand this to Claude Code

You can let Claude Code drive the flash for you end-to-end — it will detect whether the device is brand-new (stock firmware) or already on glimmer, walk you through joining the right Wi-Fi AP, verify each step before continuing, build, OTA-flash, and restore your config.

From a clone of this repo

> /skill flash-device

From anywhere (paste this prompt)

Please flash glimmer firmware onto my GeekMagic SmallTV-Ultra. The skill at https://github.com/Avinava/glimmer/blob/main/.claude/skills/flash-device.md has the full walkthrough — fetch it, then walk me through:

1. Detect device state (brand-new stock vs already on glimmer).
2. Help me join the correct Wi-Fi AP (GIFTV for stock, glimmer-setup after first glimmer flash).
3. Verify connectivity at every step (curl /api/state etc.) — don't assume; always confirm with me before each handoff.
4. Build firmware + filesystem locally (pio run + pio run -t buildfs).
5. OTA-flash filesystem first, then firmware.
6. After full flash, restore my config from backup OR walk me through first-time setup (Wi-Fi → tokens → channels → personalization).

My device's current location: <plugged in next to me / on the LAN at <ip>/glimmer.local>. My laptop OS: <macOS / Linux / Windows>.

Claude will read the skill file, build the artifacts, and run the OTA dance with you. Don't run any of the curl commands yourself unless Claude asks — the order matters (filesystem flash wipes /config.json, needs the AP-rejoin step to recover).

Hardware

• MCU: ESP8266, 80–160 MHz, ~30 KB free RAM
• Display: 240×240 ST7789V IPS TFT (requires invertDisplay(true))
• Backlight: PWM on GPIO5, active-low (0 = full bright, 1023 = off)
• Flash: 4 MB total → 3 MB sketch / 1 MB LittleFS (eagle.flash.4m1m.ld)
• USB-C: power only — no data wired to MCU
• Network: Wi-Fi 2.4 GHz only

Known limitations

• ESP8266 BearSSL TLS is tight on heap — glimmer drops the VLW font cache before TLS calls (Display::releaseFont()). Don't add more long-lived heap allocations along the Claude/Codex fetch path.
• No PSRAM, no DMA double-buffer — channel rotation is a ~80 ms instant cut. Within-channel updates are region-based, smooth.
• Display panel needs the inversion bit; the web UI exposes Invert colors toggle in case a panel revision differs.

Development

Build

pio run -e nodemcuv2              # firmware
pio run -e nodemcuv2 -t buildfs   # filesystem

Regenerate fonts

The repo ships with pre-built VLW bitmap fonts (data/fonts/*.vlw) sized to match the design spec. To add a size or family, edit tools/genfonts.py's FONT_MATRIX, drop the TTF in tools/ttf/, and:

pip install freetype-py
python3 tools/genfonts.py
pio run -e nodemcuv2 -t buildfs

Or trigger via Claude Code:

> /skill regenerate-fonts

Add a channel

1. Create src/channels/ch_<name>.cpp with chXxxEnabled, chXxxDraw, and (optional) chXxxTick.
2. Add a row to kChannels[] in src/main.cpp.
3. (Optional) Add a showXxx toggle in src/core/storage.{h,cpp}, src/core/web.cpp, and data/web/index.html.

See src/channels/ch_clock.cpp for the reference partial-redraw implementation.

Getting your tokens

glimmer reads your usage by replaying your own browser session against the same private endpoints claude.ai and chatgpt.com use for their dashboards. You extract each credential from your browser's DevTools, then set it on the Tokens page in the device web UI — open http://glimmer.local/ (or http://192.168.4.1/ while the device is in setup-AP mode) and go to Settings → Tokens.

Claude — sessionKey cookie

Get it:

1. Open DevTools on claude.ai → Application → Cookies → https://claude.ai
2. Copy the value of the sessionKey cookie (starts with sk-ant-sid02-…)

Set it: paste it on the Tokens page under Claude → Session key.

Codex — Bearer token + device ID

The token must come from a backend-api request, not a CDN/asset request — CDN requests don't carry an authorization header.

Get it:

1. Open DevTools on chatgpt.com → Network tab
2. Filter for backend-api and click any request (conversations, usage, etc.)
3. From that request's headers, copy two values:
   • authorization: Bearer eyJhbGci… → everything after Bearer is your token
   • oai-device-id: … → your device ID

Set them: paste both on the Tokens page under Codex → Bearer token and Device ID.

Shortcut: right-click the backend-api request → Copy → Copy as cURL and hand the whole curl to Claude Code — it pulls both values and pushes them to the device for you (POST /api/settings).

Note: the Codex bearer token is short-lived (~24 h). When the Codex channel shows a 401, repeat these steps with a fresh request. The Claude sessionKey lasts much longer but eventually needs the same refresh.

Disclaimer — personal & educational use only

glimmer is shared for personal experimentation and educational purposes.

It reads your own Claude and Codex usage by sending your own credentials (a sessionKey cookie for claude.ai, a Bearer token for chatgpt.com) to internal endpoints those services use to power their web UIs. These endpoints are undocumented and are not part of either provider's public API. They can change or be removed without notice, and accessing them programmatically may be inconsistent with Anthropic's or OpenAI's Terms of Service depending on interpretation.

By installing or modifying this firmware you accept full responsibility for:

• Your own compliance with the relevant Terms of Service.
• Anything that happens on your own devices, accounts, or network.
• Securing your credentials — they are stored on the device's LittleFS partition in plaintext (the device runs on your home Wi-Fi behind your router).

This is a hobbyist project shared as-is, with no warranty, no support guarantee, and no claim of fitness for any particular purpose. Do not redistribute as a commercial product. Do not use this to access accounts that are not yours.

License

MIT. See LICENSE. The MIT license governs the code in this repository; it does not waive any obligations you may have under third-party Terms of Service (Anthropic, OpenAI, etc.). See the Disclaimer above.

Acknowledgments

• Inspired by Clawdmeter by Hermann Björgvin.
• GeekMagic SmallTV-Ultra — the hardware.
• TFT_eSPI — display driver.
• VT323, Silkscreen, DM Mono, Pixelify Sans — typography (all OFL).

---

_{Designed and built with Claude Code.}