12 Hobbies StudioOpenScribe
An open-source, self-hosted AI voice recorder in the spirit of a Plaud device. Record on a small ESP32-S3, sync to your phone and your own server, and turn recordings into transcripts and summaries with any AI you choose.
self-hosted
Bring your own AI
This is the difference. A commercial recorder locks you to its cloud. OpenScribe lets you point transcription and summaries at whatever you want - a local model, an open-standard endpoint, or a commercial API. Pick one and drop it in your server config:
Transcription is just as flexible: local
faster-whisper, or any OpenAI-compatible audio endpoint (OpenAI, Groq, or your
own whisper server).
Two ways to build it
A compact wearable, or a dev board for the best audio. Same firmware, same app.
Wearable build
A XIAO ESP32-S3 with onboard mic and microSD in a printed pebble case. Pocket or lanyard. The smallest way to carry it.
Dev / kit build
An ESP32-S3 devkit, an INMP441 I2S microphone, a microSD and a LiPo. Easiest to probe, and the cleanest sound.
What it does
The expected recorder features, without the lock-in.
One-button recording
Capture to microSD on a wearable ESP32-S3. No length limit.
Three ways to sync
BLE for control, WiFi to the app, and independent WiFi upload when on charge.
Transcribe + summarise
Server-side, with the AI provider you chose. Overview, key points, action items.
Completely open API
List, download and export recordings (audio, TXT, SRT, VTT, Markdown, JSON).
Self-hosted storage
Upload to your own S3-compatible store, WebDAV or NAS. Your data stays yours.
Mobile app
Flutter app for Android and iOS: provisioning, library, playback, summaries.
Build one
Off-the-shelf parts, a printed case, and a server you run.
1. Get the parts
ESP32-S3 with PSRAM, an I2S MEMS mic, a microSD card, a LiPo and a charge IC. Full BOM in the repo.
2. Flash it
PlatformIO builds two board profiles. Record to WAV, then sync over WiFi and the on-device REST API.
3. Print the case
Parametric OpenSCAD enclosure you can re-tune to your exact board and battery.
4. Run the server
FastAPI ingest + transcription + summaries. Point it at your chosen AI and storage.
Open all the way down
Firmware, server and app are GPL-3.0. The hardware design is CERN-OHL-S. The case and docs are CC-BY-SA. Anyone can build it, improve it, and keep the result open. OpenScribe is an independent project inspired by the Plaud class of device; it is not affiliated with Plaud.