LilL3x: AI, Your Way

The search for the perfect AI chatbot continues. Everyone seems to agree that text-based interactions in a web browser are not the best path forward. But beyond that, opinions vary wildly. One of the most promising options at this time is the dedicated voice assistant. What could be more natural than chatting with an AI in the same way that we talk with other people?

There are still very few decent commercial AI voice assistants out there, so hardware hackers have been hard at work making their own. The latest to come our way was created by Kimberley Gray. It is a desktop chatbot called LilL3x, and it is made entirely from accessible hardware, so it is simple to reproduce. LilL3x also has some unexpected touches that make it well worth checking out.

The system is built around a Raspberry Pi 4 Model B, paired with a Seeed Studio ReSpeaker 2-Mics Pi HAT to enable far-field voice capture. The device combines a microphone array, a small speaker, and a Pi Camera Module to create something closer to a digital presence than a basic assistant. Housed inside a 3D-printed enclosure with a tiny OLED display, it sits on a desk like a little robot — complete with an animated face.

Rather than locking users into a single AI model, the system supports multiple large language models, including cloud-based and local options. Users can configure it to work with services like ChatGPT, Claude, Gemini, or a locally hosted Ollama instance. This modular approach extends to speech synthesis as well, with compatibility for engines such as ElevenLabs, Amazon Polly, and others.

Interaction is handled through a wake-word system powered by tools like PicoVoice Porcupine or Vosk. Once activated, LilL3x records audio, processes it through speech-to-text software, sends it to the chosen AI backend, and then responds using text-to-speech. While technically a pipeline of separate systems, the experience is designed to feel conversational and immediate.

With the camera, LilL3x can detect when someone is nearby and capture images to provide contextual awareness. In practice, this allows the assistant to check in with the user, initiating interactions rather than simply waiting for commands. It’s a small feature, but one that pushes the device closer to feeling like an actual companion.

Builders will need to assemble hardware components, configure Linux-based software, and manage API keys for various services. However, the project provides a streamlined installation script and a web-based configuration interface, making it far more approachable than many similar builds.

By giving AI a physical form — even a simple one — it changes how users relate to it. A voice coming from a device with a face, however crude, feels more engaging than disembodied text on a screen. If you’d like to experience it for yourself, be sure to have a look at the build guide.