The Open-Source Intelligence Frontier.
"We have Personal Computers — to create, work, and produce things that matter. We have Tiny Supercomputers — Smartphones — to consume, but we are drawn into the Social Dilemma.
But why have humans never built something different?
They tried. Ten years before smartphones, they made wearables — but added a screen to it. A screen that demands attention. A screen that distracts. A screen we never needed.
That is why we are building Aura.
A device that combines the productive power of a computer and the connectivity of a smartphone — but with one fundamental difference. It does not pull you in. It works for you, quietly, in the background — transforming your intent into action, without ever asking for your attention.
Less distraction. More done."
AURA is a wearable AI context engine designed to capture, process, and remember environmental signals in real-time. It transforms your daily life into a searchable, semantic memory—controlled entirely by you.
Deploy Backend • Flash Firmware • Get the App • Issues
Fits in the palm of your hand. |
Packed with intelligence. |
AURA leverages the compact power of the XIAO ESP32-S3 Sense to pack high-fidelity sensors into a form factor smaller than a coin. Unlike proprietary devices, AURA gives you the Raw Stream—no gatekeepers, no subscription fees.
- Continuous Context: Passive capture of audio and visual frames.
- Privacy First: Local streaming + self-hosted RAG backend.
- Extreme Portability: 350mAh LiPo powered, 3D-printable enclosure.
Every component is off-the-shelf. Total BOM cost: ~$60 USD.
AURA doesn't just store files; it understands them. Our Multi-Threaded RAG (Retrieval-Augmented Generation) pipeline processes data through three distinct layers:
- STT: Deepgram Nova-3 for 98%+ transcription accuracy.
- Vision: GPT-4o Vision analysis of environmental frames.
- Streaming: Efficient MJPEG and PDM audio over WebSockets.
- Vector DB: Pinecone-backed semantic indexing.
- Person-Awareness: Intelligent speaker identification and assignment.
- Contextual Chunking: Multi-threaded extraction of relevant conversation segments.
- Searchable Life: Instantly recall "The meeting details from Tuesday."
- Agentic Chat: A personal AI that knows your world as well as you do.
| Use Case | How AURA Helps |
|---|---|
| Professional | Instant summaries of every meeting, searchable by speaker or topic. |
| Academic | Never miss a lecture detail; ask questions to your class recordings. |
| Creative | Capture "Eureka" moments as they happen, indexed by visual context. |
| Personal | A searchable archive of family stories and shared memories. |
Front & back. Camera lens + magnetic clasp. |
Open internals — ESP32-S3 Sense module + LiPo. |
Exploded CAD view — 3D-printable, fully open. |
AURA is designed for Assembly. Every component is off-the-shelf, making it the most accessible AI wearable on the planet.
| Component | Specification |
|---|---|
| MCU | Dual-core 32-bit LX7 @ 240MHz |
| RAM | 8MB PSRAM + 384KB SRAM |
| Camera | OV2640 2MP (1600x1200) |
| Mic | Digital PDM Microphone |
| Wireless | 2.4GHz WiFi + BLE 5.0 |
- Phase 1: Foundation — Core streaming, transcription, and basic RAG.
- Phase 2: Vision 2.0 — Real-time object detection and spatial awareness.
- Phase 3: Multi-Modal Agents — Proactive AI assistance based on current context.
- Phase 4: Swarm Intelligence — Multi-device context sharing and collaborative memory.
AURA — Open Hardware · Open Firmware · Open Intelligence
â� Star on GitHub · Issues · MIT License
