Size
70 × 37 × 104 mm
Weight
~250 g
Battery
2 × 18650
Speaker
40 mm
Board
ESP32-S3
Print time
~6 h
What it does

Everything by voice.

HX-1 is the microphone and the speaker. A Huxley server at home does the rest: the persona, its skills, and OpenAI's realtime voice model.
  1. 01 · Calls

    Contacts call it on Telegram and it picks up by itself. Say a name to call them.

  2. 02 · Messages

    New messages are read aloud as they arrive. Reply by voice.

  3. 03 · Books, music, radio

    Audiobooks and music from a Spotify Premium account, and live radio.

  4. 04 · Reminders

    If nobody confirms a medication reminder, it repeats three times: 5, 10 and 30 minutes apart.

  5. 05 · News and answers

    Headlines, the weather, and whatever a web search can answer.

  6. 06 · What it needs

    Wi-Fi, a computer that stays on and an OpenAI API key. Calls need a spare SIM for Telegram; books and music need Spotify Premium.

About personasThe firmware ↗
Make one

Print one. Give it to someone.

About six hours on a Bambu P2S, about $50 of parts per unit, some soldering, glue and eight screws. The build guide shows all 21 steps in 3D, with print files for any printer.

Case, firmware and server are MIT-licensed. Whoever builds it runs the server and edits the persona, a YAML file; it ships with Abuelo, slow and warm, in Spanish. Whoever uses it only holds the lever. OpenAI charges about $4 per thousand answers, and nothing while idle.

The build guideThe enclosure source ↗