The ARIA Node Is Open Source. Go Build One.
- Rich Washburn
- 2 days ago
- 3 min read
I spent the weekend with it. Really lived with it — pocket, desk, driving, walking. The voice recorder mode. Push, talk, done. And I kept thinking: this is actually it. This is the thing I wanted.
So I'm giving it away. Here's where things landed and why.
Two Versions. One Winner.
If you've been following this build, you know there were two paths. The S3 — full display, menu carousel, mode selector, name on the screen. And the Nano — the XIAO ESP32-C6, stripped down, no display, just push-to-talk in the smallest possible package. The Nano was a constraint exercise. Interesting engineering. And ultimately, not ready — the C6 just doesn't have the chops for what this firmware demands. The right board for that form factor is the S3, same as the display version. When I get around to revisiting the Nano concept, it'll be on an S3 in a smaller enclosure, not a different chip family. For now, the C6 version is shelved.
The S3 display version? That's the one. It's finished the way a thing feels when all the rough edges are gone. Every button press is intentional. The carousel scrolls cleanly. Dictation mode works. I've been using it all weekend and I haven't wanted to put it down.
What You're Getting
This is the full ARIA Node v2.5 source code. Open source, no strings attached.
Four modes on the carousel:
Dictation — fully offline WAV recording to SD card. No WiFi, no cloud, no dependencies. This is the voice recorder mode. Press, talk, done. Files save locally and you pull them off via USB or SD card.
ARIA Agent — WiFi streaming to your AI endpoint for live interaction. The full loop: press, speak, Whisper transcription, agent response on screen. No phone. No laptop.
USB Drive — the device bridges the SD card to USB. Your computer sees a flash drive. Drag your recordings off directly. No adapters, no apps.
Setup Portal — broadcasts its own WiFi network. Connect from your phone, a config page appears automatically. Set your bot name, colors, WiFi credentials. Saves to config.json on the SD card. No IDE required to configure it.
The hardware is under $30 all-in on an ESP32-S3 with a Waveshare 1.47" display, an INMP441 MEMS mic, microSD, and a 3-button rocker. Full BOM and wiring notes are in the source.
One thing to be clear on: this is source code, not flashable firmware. You'll need to open it in the Arduino IDE, wire up your libraries, and compile it for your board. If you don't know what that means yet, there are about a thousand YouTube videos that will get you there in an afternoon. It's not hard. You just have to actually do it.
Why Open Source
Because the interesting part to me isn't the device anymore — it's using the device. I've gotten it to the place where it does what I wanted it to do. The dictation workflow alone is worth the build. The agent mode is the next chapter.
I'd rather see ten people building their own versions and pushing it in directions I haven't thought of than sit on it.
The pattern this represents — a physical, wearable, offline-capable interface to your own AI — that's where things are going. This isn't a hobby anymore. It's the leading edge of how AI gets embedded into the physical world. I built mine from the pocket up.
Get the Source
The .ino (.txt) source file is linked below. Grab it, open it in Arduino IDE, make it yours. Swap the AI endpoint, rename the device, change the colors, redesign the whole thing — go for it. And if you build something awesome with this — something cleaner, smarter, better enclosure than my janky 3D-printed version — genuinely, I want one. Send me a link. Or just send me the thing. I'm not above accepting tribute from projects I accidentally started.
Hardware used: ESP32-C6-LCD-1.47
— Rich
Comments