Made a little device that manages my claude code’s & codex’s on my server… Simple esp32 s3 chip by waveshare. The orchestrator agent works with voice, controls my device remotely with tools, and controls the server. It’s a cool time to make stuff.

Hey, I built this! Nice work ♥️

Looks cool!

Some questions: - The esp32 is WiFi connected right? No SIM card? - VoIP hosted? - TTS / STT Models hosted or are you using a provider? - What’s the latency when you are walking outside? - How long does the battery last when you are walking around talking?

Yup, it’s wifi (but that chip has bluetooth too) VoIP on the server, experimented with open TTS/SST/LLM on server it connects to, but also all eats resources/compute I want to use to do work etc. Outside it uses phone as wifi hotspot, so same latency

Any reason why you won’t just use your phone? Self host FreePBX, add a SIP-trunk so you’ve got phone numbers for your agents.

What kind of hardware is that? It’s not only the esp32-s3 , I really like the integration. How can I build same?

same waveshare s3 sitting in my parts drawer for 6 weeks unopened

This is very cool. A custom piece of hardware whose only job is to boss your agents around would've sounded insane two years ago.

You don't need a separate devices for everything... Just make an iPhone / Android app that has single "Voice" button. You could use it and on top of that, you could sell it on mobile stores.

This looks really cool!!

Thanks Daniel!

Next level: add local STT with Whisper-tiny + edge vector DB (like Chroma or LanceDB) so it can pull context without hitting the server every time.

Neat :) What's the voice agent btw? And did you write the UI etc for the device or is that using an existing project?

Voice agent is gemini live sdk, but could probably use any of em, eleven labs etc. For sure I had to write that UI, but claude helped me a ton. Also wrote the games and all the entire firmware on the device

Very cool, Louie! 😎 I was hoping to build a personal assistant but local. Maybe it'll be possible now

Impressive design and ui too ! Any primers on how to get started with custom enclosures for hardware ? I have a 3d printer btw

This is exactly the kind of thing I would build and then immediately ask myself why I gave it server access.

This is sick dude I’m going to order one to mess with

I want this to control my brainstack which already uses telegram to control a fleet of servers with yolo codexes on them. This is obv the next upgrade

Was eyeing the S3 just the other day. Seeing the 240x240px resolution has me realizing it's a lot crisper a display than I was expecting.

Brooooo this is the coolest shit 🥹🥹🥹

thanks your inspiration I made one as well 😂

So cool! Would love to learn more about this.

I love this. Reason being the idea is not unique. The execution seems super clean. Voice is fast and natural sounding. The listening button is perfect. If you have project files or love to contribute to the git.

This is what OpenAI were going to build, haha

This is so nice 👌🏼

This is really cool, Louie!

😆 I have the same device — just doing something a little different with it.

This is the coolest version of an itch a lot of us have — remote-controlling the agents on your own box. I scratched the same one in software: Telegram as the remote. Text it a spec, it builds + deploys to my VPS, texts back the live link. Yours is way more fun to look at though.

wow. looks really great 👍

I’ve been doing something similar but this looks better nice work

Beautiful

The S3 is a beast for how cheap it is. Physical control loops for server agents actually make sense if you want a quick way to monitor them without opening another browser tab.

That’s super cool. Adding a screen would make it even cooler.

Nice work! Looks pretty good. What did you use for voice sythesys?

@daniel_nguyenx This is so cool. I want one now 😬

Nice one! Release that UI on GitHub ...star bound.

really cool!

I really like physical extensions for coding Agents.. Bravoo