dream-mosaic/orbital-pai

Orbital P.A.I is a voice controlled personal assistant that you talk to out loud and that talks back, aiming to respond fast enough to feel like a natural conversation rather than a slow chatbot. It is built with Elixir and Phoenix, a web framework often used for real time applications. Beyond just chatting, it can actually take actions for you, such as checking the weather, managing reminders, reading and sending email, and working with your calendar, all through what the project calls a tool calling brain, meaning the AI can call specific functions to get things done mid conversation. The assistant listens through your microphone, converts your speech to text, and splits its response into two parts: a quick instant filler reply that covers the short delay while a fuller, more considered answer streams in and is converted back into spoken audio. This two step approach is meant to make waiting for a response feel less noticeable. It also remembers useful facts about you across separate conversations, supports natural pauses in speech without cutting you off, lets you interrupt it while it is talking, and can speak up on its own to deliver a reminder at an appropriate moment. It is designed to work as an installable web app on a phone as well as in a dedicated always on kiosk mode, useful for mounting somewhere like a wall or a desk. Multiple people can use it, since sign in is limited to an approved list of users and each person's memory, reminders, and connected accounts stay separate from everyone else's. Running it requires an Elixir and Node.js development environment along with paid API keys from a few outside services for the AI language model, speech recognition, and text to speech, so there are ongoing costs to running your own instance. Optional Google account integration for calendar and email needs its own one time setup. The project is explicitly under heavy active development, so the author warns that bugs and breaking changes should be expected. It ships as a Docker image for deployment and is released under the MIT license.