# I Built a Minimalist AI Note Device

[Watch on tuber](https://tuber.guzus.xyz/video/3t0k7E7WiOQ/summary) · [YouTube](https://youtu.be/3t0k7E7WiOQ)

**Score: 64/100 · skim**

The creator built the **Palma Note**, a small pocket voice-note device. It runs on a Waveshare ESP32 board with an e-ink display and is meant as a "second brain": you can capture an idea in a few seconds without unlocking your phone and getting distracted. Recordings are saved to an SD card, sent to OpenAI Whisper for transcription when the device syncs over Wi-Fi, and can be exported through a local web interface. The files, firmware and assembly guide are on the creator's Ko-fi page.

- **0:02-0:54** — **The problem and the parts list.** Ideas disappear within seconds if you don't write them down. Phones are poor for this because unlocking one to jot a note usually leads to distraction. The creator lists what a dedicated device needs: a microcontroller, display, microphone, speaker, battery, storage and good power management. He admits that's "more than simple," but a single Waveshare board turned out to include nearly everything.

- **1:00-1:55** — **Why ESP32 and e-ink.** The ESP32 is a popular low-power microcontroller with built-in Wi-Fi and Bluetooth, and it's fast enough to run the device logic. An e-ink screen only uses power when it updates, so a static image stays on screen without draining the battery. It's also calmer to look at than a bright phone screen. He picked parts that are easy to buy and assemble, so beginners can build the same device. All files, firmware and the assembly guide are on Ko-fi.

- **2:02-3:09** — **Custom snap-fit case.** The board ships with its own enclosure, but he rejected it as too boxy and limited to specific battery sizes. People had enjoyed modding his earlier "tiny e-reader" project, so he designed a case that is easy to print and customize, with as few extra parts as possible and **no screws**. After many prototypes, the two halves now snap together with side clips. Four internal pins hold the board in place, the buttons drop into cutouts, and almost every edge is rounded so it's comfortable to hold and tough enough for a pocket or backpack.

- **3:36-3:55** — **Battery choice.** He used a 500 mAh LiPo to keep the device compact. Larger batteries would fit if you enlarge the case slightly. He expects the community to extend the design well beyond this first version, as happened with the e-reader.

- **3:59-4:19** — **Development was harder than expected.** He delayed the video several times to fix bugs, mainly in **audio recording and power management**. He thought it would take a few evenings, and it took far longer. He's only now happy with how it works.

- **4:19-5:07** — **The interface.** The device spends most of its time in an ultra-low-power deep-sleep screen. Holding a button wakes it to a home screen, where a ring around the display shows the battery level. You record by holding the upper button and talking; his demo note is "don't start another side project." Releasing the button saves the clip, plays a confirmation sound and opens a tag menu (e.g., "project"). After you pick a tag, it goes straight back to deep sleep. The whole interaction takes a few seconds, which is the point.

- **5:07-5:59** — **From voice recorder to the missing piece.** Even this version works well as a small voice recorder: clips go to the SD card and play back through the built-in speaker. But a large pile of audio files is hard to search, so he wanted transcription. An ESP32 can't realistically transcribe speech on the device, so after trying a few options he chose **OpenAI Whisper** in the cloud.

- **6:01-6:48** — **Setup and syncing.** Before flashing the firmware, you enter your Wi-Fi credentials and OpenAI API key in the code's config section. To save battery, Wi-Fi doesn't run in the background. The device syncs automatically **once per day**, or you can start a sync from the menu. It uploads recordings one at a time, and Whisper's text files are saved back to the SD card, so every note exists as both audio and text. Note that this depends on a paid OpenAI API key and sends your recordings to the cloud.

- **6:48-7:08** — **Ideas for future AI features.** Because the device already talks to an AI service, later firmware could summarize notes, generate titles, categorize ideas or create daily overviews. These are possibilities he mentions, not features that exist yet.

- **7:08-7:51** — **Transfer mode for your own workflow.** To get notes into tools like Notion or Obsidian, transfer mode connects the device to your home network and shows a local IP address on screen. Typing that address into a browser opens a minimal web interface. There you can browse notes, filter by tag, download text files and export the audio recordings.

- **7:51-8:18** — **Small quality-of-life feature: button sounds.** At first he thought button sounds were unnecessary. But e-ink refreshes slowly, and a short sound makes each press feel responsive and tactile, which makes the device feel more alive.

- **8:18-8:45** — **Wrap-up and question to viewers.** He asks what features people would add next. He also asks whether parts of the tiny e-reader firmware, which people have started building apps for, should be ported to the Palma Note, or whether the screen is too small. The rest is the usual sign-off.

**Practical takeaways:**
1. A single-purpose capture device avoids the phone's distraction problem. Recording, tagging and returning to sleep takes a few seconds.
2. Pairing e-ink with ESP32 deep sleep is a good pattern for battery-powered devices that are mostly idle. Turning Wi-Fi on only during syncs saves even more power.
3. Hand heavy work like speech-to-text to a cloud API (Whisper) instead of trying it on the microcontroller. Plan for the API cost, the privacy tradeoff and the need for Wi-Fi.
4. If you build it, expect audio recording and power management to be the hardest parts. A snap-fit, screwless printed case makes it easy to modify, for example to fit a bigger battery.
5. If you want your notes in Notion or Obsidian, look for a local web export (like this transfer mode) rather than relying only on files stored on the device.

## Audience signal

- No high-signal audience themes available.