Inspiration Luna was inspired by Tamagotchi and the idea of having a small AI buddy that lives with you, not inside your phone. We wanted quick and easy access to AI without always opening a screen, scrolling, or getting pulled into apps.
What it does Luna is a daily AI buddy and voice-first agent. She lives on a Raspberry Pi with a small screen and voice control, so you can talk to her naturally from the palm of your hand.
You can use Luna for daily chats, emotional support, reminders, idea building, web search, or more complex tasks. She can search knowledge on the internet and remember what you have told her through local memory. Her supportive personality makes her feel like she is always there with you when you need help. She can also self-evolve and only apply her new skill when you approve.
How we built it We built Luna on top of a Raspberry Pi with a screen, speaker, microphone, and voice controls. The device connects to a laptop bridge that handles local memory, internet search, project context, and AI orchestration.
We built an orchestration layer that routes each user utterance to the model or tool that fits the context best. Luna uses DeepSeek and MiniMax hosted on CLŌD for fast model routing, with the OpenAI API as a backup when needed. This helps Luna respond quickly and reliably, whether the user is asking a daily question, looking up information, having an emotional conversation, or starting a complex task.
Challenges we ran into The hardest parts were model orchestration, memory, web search, communication misfires, and unstable hardware networking.
Luna has to understand what the user really means, then decide whether to use memory, reasoning, web search, or another model. Sometimes Luna misunderstood casual phrases, like treating “Have a good day” as a question about today’s date because she focused too much on the word “day.”
Memory was also tricky because Luna sometimes relied on saved memory too quickly, even when she needed to reason more or search for current information.
We also ran into hardware and networking instability. Since Luna runs on a physical Raspberry Pi and connects to a bridge, connection drops, Wi-Fi changes, and device setup issues could interrupt the experience.
Another big challenge was keeping Luna user friendly. Since Luna is voice-first, she cannot give long responses full of jargon. Her answers need to be short, clear, and easy to understand.
Accomplishments that we're proud of We’re proud that Luna feels like a real buddy, not just another chatbot. She can talk, listen, remember context locally, search the internet, give support, and help with bigger tasks.
We’re also proud that we built a working AI companion on real hardware, with an orchestration layer that can use multiple models instead of depending on only one provider.
What we learned We learned that a daily AI buddy needs more than intelligence. It needs good timing, short responses, emotional warmth, memory, and the ability to choose the right tool or model at the right moment.
We also learned that building AI on hardware adds real-world complexity. Networking, device setup, latency, voice control, and recovery from connection issues are just as important as the AI model itself.
What's next for Luna - Your daily AI buddy Next, we want to make Luna faster, more stable, and more emotionally aware. We want to improve local memory, web search, intent detection, model routing, and hardware networking.
We also want to make Luna feel even more alive with better animations, smoother voice interaction, and stronger support for daily companionship, emotional support, internet knowledge search, and complex tasks.
Built With
- clod
- codex
- cursor
- node.js
- python
- raspberry-pi
Log in or sign up for Devpost to join the conversation.