Say Alexsay
Say the name and it is listening. The wake word runs on the speaker itself, so nothing leaves the room until you talk to it.
Open-source home assistant speaker
A speaker that listens to you.On your terms.
Alexsay is a voice assistant you build and own: a Raspberry Pi 5, a four-microphone array and the AI provider you choose, on your own key.
Hardware, firmware and this site are open. Your API keys stay yours.
Built for conversation, not commands.
Everything that makes a speaker pleasant to talk to, with nothing hidden inside.
Talk, do not take turns
Full-duplex voice. Interrupt the assistant mid-sentence and it stops and listens.
Hears you over itself
A four-microphone array with hardware echo cancellation keeps its own voice out of your words.
Your keys, your bill
Bring your own OpenAI key today, Claude and Gemini next. Nothing is resold and your keys stay encrypted.
Runs your home
Lights and more through the services you connect: Lightwave today, Apple Home and Google Home next.
Open hardware, open source
A printed enclosure, off-the-shelf parts and a Rust daemon you can read, build and change.
Plugs into what you already use.
Pick an AI provider for the conversation and the smart home it should control. More providers are on the way.
OAOpenAIGPT-Live voice
CLClaudeAnthropic
GEGeminiGoogle
AHApple HomeHomeKit
GHGoogle HomeMatter
LWLightwaveLights and sockets
HAHome AssistantLocal hub
Made to be heard, not looked at.
A seamless fabric wrap around a 4 inch driver, a four-microphone array under the top. No screen, no buttons: you talk to it, and it talks back.
- Raspberry Pi 5 inside
- Hardware echo cancellation
- 3D-printed enclosure
- Mains powered, one cable
Open hardware. Open source.
The parts list, the printed enclosure, the Rust daemon, the app and this website live in one repository. Read it, build it, change it.
Development
Install it in one line. Talk to it in the next.
The speaker is one Rust binary. Download it, train the wake word on your own voice and say its name.
curl -fsSL https://alexsay.com/install.sh | shBlazing-fast Rust
One static binary under 13 MB for Linux, the Raspberry Pi, macOS and Windows. It starts in a fraction of a second.
Nothing to install
Static musl on Linux with the audio stack compiled in. No Python, no Node.js, no runtime, no packages: drop it in and run it.
Wake word on the device
A small neural network listens for its name locally. Record a few takes and train it on your own voice in minutes.
Echo cancellation in software
WebRTC's AEC3 in pure Rust keeps the speaker's own voice out of the microphone, so you can talk over it.
Your own server in one file
alexsay-server carries the API, the app, an HTTP server and an embedded SQLite database. Same code as the cloud.
Signed updates
Every release is built and run on its own architecture, signed, verified by the speaker and announced by voice.