Crab StickConversations you can see
Read captions on the device. Follow live transcripts, streamed replies, stages and errors in your browser.
YOUR DEVICE. YOUR MODELS.
A little companion,\non your terms.
Hold to talk. Release to listen. Turn your StickS3 into a pocket voice companion, powered by Pi on your computer.
No Muse account or subscription · Your choice of model and speech
A SMALL FRIEND.
A REAL CONVERSATION.
A REAL CONVERSATION.
CRABWiFi ●
Hello, I’m Crab.
Your voice, a conversation away.
VERIFIED ON HARDWARE
StickS3 + macOS + faster-whisper + Pi / DeepSeek + macOS TTS
SMALL HARDWARE. OPEN POSSIBILITIES.
A Crab in your hand.\nA brain on your computer.
The device records, animates, displays captions and plays audio. Your computer handles speech and model access. API keys stay on the computer.
01Hold to talkStickS3 · 16 kHz WAV
→
02Recognize & thinkComputer · STT + Pi
→
03Read & listenCaptions + PCM audio
Bring your own model
Select configured providers and models through Pi. Cloud usage follows provider pricing; local models need your own setup.
Make it sound like you
Tune Whisper, recognition language, the installed voice and speech rate. Or switch to text-only replies.
FROM THE WORKBENCH TO YOUR POCKET
Ready to meet your Crab?
1. Start the bridge
Run from the repository root. Keep the terminal open and computer awake. Run npm run build first after code changes.
cd crab/bridge
npm start2. Open your local console
Open the address below and sign in with CRAB_TOKEN from your local crab/bridge/.env. Switch between English and Chinese in the header.
http://127.0.0.1:8787/ ↗Works on the computer running the bridge. This project website is separate from your device console.
03 /
Hold the StickS3 front button, say “你好，你是谁”, and release. Check the captions and spoken reply.
You’ll need a StickS3, USB data cable, reachable 2.4 GHz Wi-Fi, Node.js ≥ 22.19, Python 3 and ESP-IDF 6.0.1. The verified local speech setup uses macOS.
-
Get the code & install dependencies
git clone https://github.com/everettjf/crab-stick.git cd crab-stick/crab/bridge npm ci --ignore-scripts npm run build cp -n .env.example .env -
Configure the computer
Generate a shared token, configure DeepSeek / Pi and local speech, and start with demo mode. Follow the linked setup guide for the full configuration.
-
Back up before flashing
Save the full 8 MiB flash and SHA256, then build and flash Crab. Configure Wi-Fi, the computer’s LAN address and the same token over USB.
-
Go from demo to real conversation
Verify fixed captions and the demo tone, then set CRAB_DEMO=false and restart the bridge. Test recognition, captions and speech near the microphone.
A FEW BUTTONS. THAT’S ALL.
Keep it hands-on.
FRONT · HOLD
Record. Release to send.
Up to 20 seconds per turn.
FRONT · PRESS WHILE BUSY
Cancel the current turn.
Start again once cancellation finishes.
SIDE · SHORT PRESS
Turn the page. Make it happy.
Page through captions and trigger a happy reaction.
A WORKING FIRST VERSION
Know what’s ready.
Chinese recognition, captions, voice replies, cancellation, reconnection and heartbeats have been exercised on hardware. Battery runtime, battery-only wake and long-term stability still need measurement.
Read the validation record ↗A few things to remember
- Keep the computer awake and reachable at its LAN IP.
- Chat only: no reminders or persistent notes. Restarting the bridge clears conversations.
- Use HTTP on a trusted LAN. Do not forward the bridge port to the internet.
- Model switches clear context. Speech settings apply to the next turn.