Status How it works Specs Open source Japanese GitHub

A pebble that remembers everything.

It remembers the conversation you had at breakfast and reminds you before your afternoon meeting. 5 grams. No screen. Your memory, backed up.

~0g
TARGET WEIGHT
~0h
ESTIMATED BATTERY
Open
HARDWARE
See project status Target specs

Project Status

Koe is an open-source project in active development

Hardware
Schematic designed, components being ordered
70%
Firmware
In development (ESP32-S3 + Rust)
40%
Software
Koe voice app available for macOS/Windows — app.koe.live
100%
Prototype
We're designing the first prototype
20%
Guitar to multiple Koe devices via iPhone, 30ms latency
30ms ULTRA LOW LATENCY

Play your guitar.
Every Koe in the room plays along.

Plug a cable into your iPhone. Audio streams to every Koe device on the same WiFi in 30ms. No rehearsal studio needed. Your living room is the stage.

Signal flow: Guitar → iPhone → WiFi → Koe devices
Koe 5 models

Pick · Ear Cuff · Coin · Band · Lantern

Concept renders — prototype in progress

Ear Cuff worn on ear — mic hole and LED detail
Wear it

Tuck it behind your ear. Forget it's there.

At 25 x 30mm and just 5 grams, Koe is smaller than an AirTag. Clip it, pin it, or hang it from a chain. It records everything you say and hear. No buttons to push. No screen to check.

Microphone: INMP441 MEMS, 61dB SNR, I2S digital
Voice detection: On-device VAD, triggers only on speech
Coin device held between fingers
Hold it

Feels like a smooth river stone.

Matte ceramic finish. Rounded edges that fit naturally between your fingers. One LED that breathes softly. No logos. No labels. Just a quiet piece of tech that feels like jewelry.

Body: PC/ABS matte with ceramic-feel coating
LED: Single RGB, 7 states, ambient breathing animation
Button: Recessed side button for mute / manual trigger
Band v2 at festival
Listen to it

When it talks back, you want to hear it.

Koe STUDIO adds a 28mm full-range driver. Place it on your desk for meetings, or clip it to your bag for outdoor conversations.

Driver: 28mm full-range, 4ohm, 3W (target)
Volume: ~85dB @ 1m estimated, with DSP echo cancellation
Battery: 2000mAh target, ~24h recording estimated
Lantern STAGE at festival
Soluna Mode

Speak into one. Everyone hears it.

Medium-press the button to switch from Koe mode (AI assistant) to Soluna mode (P2P audio mesh). At a festival, conference, or any event, multiple Koe devices on the same WiFi form an instant audio channel. No server. No app. No setup. Just speak.

This is something no other AI wearable can do. Humane Pin, Rabbit R1, Friend pendant -- none of them talk to each other. Koe devices do.

Soluna: solun.art
Protocol: UDP multicast 239.42.42.1:4242 -- zero infrastructure
Discovery: mDNS auto-discovery on local WiFi
LED: Orange pulse when synced with peers
Latency: <50ms on local network
Switch: Medium press button to toggle Koe / Soluna mode
Soluna Edition (Festival-grade audio) →
Sound vs radio speed

Sound (yellow) vs radio (red) — radio is 870,000x faster. That's why CROWD arrives before the sound.

How Koe thinks.

Five AI agents work together in the cloud. Your device is the ear and the mouth.

1
Always listening, never uploading
A 30-second ring buffer on the device. Voice Activity Detection runs locally. Only actual speech gets sent to the cloud.
2
Transcribed and understood
Speech-to-text in the cloud. Context from your past conversations is pulled in. The AI knows what you were talking about 3 hours ago.
3
Thinks, researches, responds
Reasoning, web search, and memory agents collaborate to form an answer. Not a chatbot reply - a considered response that knows your preferences.
4
Speaks through your earbuds or its own speaker
TTS audio streams back. MINI plays through BLE earbuds. STUDIO plays through its built-in 28mm driver. Sub-second latency on WiFi.

Management Dashboard

Monitor and control every device in real-time. Sync status, battery, volume — all from your browser.

Monitor and control every device in real-time. Sync status, battery, volume — all from your browser.

Try the dashboard →

Every detail.

Three models. Same AI brain. Pick the body that fits your life.

All specifications are target specs — subject to change during prototyping
MINI BAND STUDIO
Size25 x 30 x 9mm42 x 18 x 11mm50 x 70 x 20mm
Weight~5g~25g~65g
Wear stylePendant / Clip / PinWristbandClip / Stand
MCUESP32-S3 dual-core 240MHz, 8MB Flash, 2MB PSRAM
WirelessWiFi 802.11 b/g/n + Bluetooth LE 5.0
Microphone1x MEMS2x MEMS stereo2x MEMS wide (40mm spacing)
Mic SNR61dB (INMP441, I2S digital output)
Speaker-- (BLE out)10 x 8mm28mm full-range
Frequency range--300Hz - 8kHzTBD (pending measurement)
Max SPL--~70dB~85dB @ 1m est.
DSPVAD onlyVADVAD + AEC + EQ
Battery200mAh120mAh2000mAh
Recording time~9h est.~5h est.~24h est.
Active streaming~3h est.~2h est.~6h est.
ChargingUSB-CMagnetic pogoUSB-C
Charge time~30 min~20 min~90 min
LED1x RGB5x RGB strip5x RGB strip
InputSide buttonCapacitive touchButton
Water resistanceTBDTBDTBD
ColorObsidian / Ceramic WhiteBlack / White / GreenObsidian
Price (target) TBD TBD TBD

Open everything.

Koe is open-source hardware. Download the schematics, print the case, flash your own firmware. We sell the device for people who don't want to solder.

~$12
EST. BOM COST
27
COMPONENTS
2
LAYER PCB
Rust
FIRMWARE
5
FORM FACTORS
KiCad
SCHEMATICS
GitHub repo KiCad files BOM
FAQ

Frequently asked questions

What is this?+

Koe is a tiny AI device that always listens to your surroundings and lets you recall information when you need it. One device is an AI assistant. When many devices come together, they form a synchronized audio mesh — turning a festival crowd into an orchestra.

When can I buy one?+

We're building the prototype now. The hardware is open source, so you can download the schematics and BOM to build your own today. A retail product is not yet available. Star the GitHub repo to follow progress.

Always recording — what about privacy?+

Koe runs Voice Activity Detection on-device. Only actual speech is sent to the cloud. Idle audio stays in a 30-second local ring buffer and is never transmitted. The firmware is open source — anyone can verify exactly what is sent.

How is this different from Humane Pin / Rabbit R1?+

Other AI wearables are shrunken smartphones — one device, one user. Koe gets better when more people use it. 100 Koe devices in sync become a distributed speaker system. No other wearable can do device-to-device audio. And it's open hardware.

Can I build one myself?+

Yes. KiCad schematics, BOM, Rust firmware, and 3D-printable case files are all open source. An ESP32-S3 dev board + INMP441 mic + MAX98357A amp on a breadboard gets you a working prototype for ~$40. See parts list →

TEAM

Who's building this

Y
Yuki Hamada
EnablerDAO / Tokyo
Building Elio (P2P AI), Soluna (festival audio), and chatweb.ai.
Koe is the device that connects them — AI's ear and mouth.
GitHub Web

Build it with us.

Koe is open source. Hardware design, firmware, acoustics, product design — all contributions welcome.