Inspiration
I started this project with a focused architectural goal: I wanted to build an AI-first, cross-platform system to organize my time, schedule deep work, and manage daily priorities without being held hostage by traditional, distraction-heavy smartphone interfaces. The objective was an eyes-free, voice-first environment that would keep technology in the background and eliminate cognitive clutter.
While engineering this screen-agnostic interaction model, I had a profound realization: an operating architecture built to function seamlessly without looking at a screen is, by its very nature, the exact tool a visually impaired or blind person needs every day. If an eyes-free, voice-driven system gives a sighted person their focus and time back, for someone who cannot see, it gives them their independence, spatial awareness, and safety back.
That realization transformed the project from a personal productivity engine into an ambient home ecosystem. A smartphone alone is insufficient when misplaced or out of reach. By turning a local private storage hub—the UGREEN AI NAS—into an always-on sovereign intelligence vault, the home itself becomes an ambient bridge. This ensures that sensory assistance, document reading, and perimeter safety are always accessible hands-free, with zero reliance on invasive third-party clouds.
What it does
BeaconVault connects everyday personal hardware into one calm, unified operating system: the phone in your pocket, ambient living room displays, and local RTSP perimeter cameras, all anchored by the UGREEN AI NAS.
Instead of searching through confusing grids of app icons, navigation works like walking through rooms in a house. With simple two-finger swipes and spatial audio cues, users can move naturally between their daily cockpit, focus timer, emergency radar, and AI camera without needing to stare at a screen.
- UGREEN AI NAS (Sovereign Intelligence Hub): Rather than sending personal health scans or audio thoughts to remote corporate servers, the UGREEN AI NAS acts as the private brain running local UGOS agents (
ug-filesystem,ug-document,ug-photo,ug-knowledge,ug-surveillance). It processes confidential mail, indexes voice memos into searchable vector stores, and monitors camera feeds entirely on-device. - Eyes-Free Document & Vision Inspection: When reading medication boxes, dosage labels, or utility bills, the system inspects captures locally using vision models and OCR. It immediately reads back critical instructions and warnings through natural speech or tactile feedback.
- Hands-Free Ambient Room Bridge: Using a self-hosted Model Context Protocol (MCP) server, ambient devices can trigger and control phone actions across the room:
- Finding a Lost Device: If a blind user misplaces their phone, a simple voice command triggers the system to ring a loud acoustic siren, flash the strobe light, and vibrate the phone immediately over LAN.
- Remote Navigation: Users can navigate between functional spaces (Cockpit, Settings, or Vault) completely hands-free.
- Private Home Perimeter & Visitor Triage: BeaconVault integrates local home cameras to provide immediate, eyes-free auditory descriptions of who or what is at the door when motion is detected, screening visitors and packages while keeping all security feeds strictly inside the house.
- Fair Access and Sovereign Safety: Every installation creates its own private, self-custodial cryptographic identity. Through transparent smart contracts, emergency SOS records and distress telemetry are securely preserved in a tamper-proof local blackbox audit trail that can never be altered or lost.
How we built it
BeaconVault is engineered as a modular, local-first ecosystem:
- Unified Multi-Platform Client (Flutter): We built the cockpit client application using Flutter and Dart, allowing a single codebase to run across mobile, desktop, and large ambient screens. Navigation uses smooth mathematical spring physics, so moving between spaces feels tactile and fluid.
- Declarative Skill & UGOS Agent Orchestration: We structured the hub using the official Uliya Skill Development standard (
SKILL.md). A coordinator skill delegates tasks to specialized sub-agents (file-agent,document-agent,knowledge-agent,photo-agent,security-monitor-agent). All tool calls enforce strict sandboxing and mandatory approval boundaries before mutating any files. - Local-First Persistence and Zero-Latency Wi-Fi Sync: Every piece of data is stored locally in an encrypted SQLite database using Drift. Within the home, devices discover each other silently using local UDP discovery on port 8491 and exchange messages over direct WebSockets on port 8492 (<2ms latency).
- Local AI & Multimodal Understanding: On-device models act as the digital eyes of the system. Camera snapshots and documents are analyzed to generate clear descriptions of surroundings, read printed medicine containers, or identify visitors.
Challenges we ran into
- Mobile Spatial Gestures vs. Content Scrolling: Resolving conflicts between scrolling inside lists and navigating between rooms by implementing dedicated two-finger gestures for spatial navigation.
- The Web 2.5 Learning Curve: Engineering self-custodial cryptographic key pairs directly on hardware to guarantee zero-password, sovereign accounts for users.
- Engineering an Extensible System: Designing a modular, pluggable architecture where any developer can register new accessibility coordinates and voice tools.
- Cross-Platform Realities: Prioritizing stable desktop and Android builds first, maintaining clean architectural abstraction for rapid future expansions.
- Balancing Multi-Agent Workflows Locally: Governing concurrent local agent tasks (OCR, transcription, surveillance) on NAS hardware with strict execution sandboxes.
Accomplishments that we're proud of
- Building an operating environment that can be fully navigated without looking at a screen, using natural spatial audio cues and simple two-finger gestures.
- Transforming home NAS storage from a passive hard drive into an active, intelligent assistant that reads mail, transcribes thoughts, and guards home perimeter safety.
- Achieving complete offline-first independence where every task, alarm, and setting remains functional during network outages.
- Releasing the entire platform as open-source code under the MIT license for the global accessibility community.
What we learned
- Real accessibility is not just adding large text or screen readers on top of visual apps; it requires redesigning interaction from the ground up using sound and touch.
- Local NAS hardware is the ideal home for assistive AI, eliminating monthly cloud fees and keeping intimate health and domestic data sovereign.
- The power of declarative agent orchestration standards (like Uliya and MCP) in driving hardware actuation safely.
What's next for BeaconVault
- Benchmarking on iDX6011 Pro: In Round 2, testing and fine-tuning local VLM inference speeds and multi-agent concurrency directly on physical UGREEN AI NAS hardware.
- High-Contrast Ambient Reading: Expanding big-screen TV and desktop modes into comfortable, high-contrast document and prescription reading.
- Community-Driven Spatial Extensions: Providing open documentation and templates so developers around the world can build new assistive plugins.
