Inspiration

Traditional travel systems offer static packages. We wanted to build something that bends the world to the human's will. No human agency in the world can research millions of live data points in seconds to build a microscopic, step-by-step reality tailored exclusively to one person. We envisioned an AI that doesn't just chat, but orchestrates life.

What it does

Xolvia OS is a hyper-personalized, autonomous life-experience orchestrator. It doesn't just chat; it acts. If a user says, "I want to go to Venice, but I am in a wheelchair, I hate the rain, and I want to eat authentic pasta," Xolvia dynamically scans global weather, verifies the wheelchair accessibility of Venetian streets and specific restaurants in real-time via Google Search Grounding, discards invalid options, and constructs a flawless, bespoke itinerary.

Beyond just planning, Xolvia features:

  • Conversational AI: Features Voice-in and Voice-out capabilities using the Web Speech API for a frictionless, human-like concierge experience.
  • Autonomous Action-Taking: It goes beyond generating text by autonomously creating .ics calendar files and dispatching premium, corporate-styled HTML emails via Nodemailer.
  • Real Visual Grounding: Extracts landmark keywords and fetches 100% real, high-resolution images via the Wikipedia API. No placeholders.
  • Smart Intent Detection: Intelligent UI that auto-downloads files when the user implies a "save" or "download" command, eliminating unnecessary clicks.

How we built it

We built a 100% Database-Driven Autonomous Multi-Agent Architecture using Google Cloud Firestore. Our system uses two main agents powered by Gemini 3.5 Pro:

  1. Router Agent: Cleans data, translates entities, and handles unclear intent.
  2. Taskmaster Agent: Takes validated intelligence, applies dynamic rules fetched via RAG from Firestore, and performs Google Search Grounding to verify real-time constraints (weather, closures, accessibility).

The backend runs on Node.js/Express, and the frontend is built with React and Vite. We also implemented Enterprise-Grade Security by encrypting all API keys and cloud configurations using AES-256 via @dotenvx/dotenvx, ensuring zero plain-text secrets in our codebase. Furthermore, we integrated Server-Sent Events (SSE) for real-time streaming to show exactly what the agent is "thinking" and researching in the background.

Challenges we ran into

Shifting from hardcoded prompts to a completely database-driven (Firestore) architecture was challenging. We had to build a robust system where agents fetch their persona, rules, and constraints dynamically at runtime. Additionally, mastering Agent Handoffs—passing unstructured, messy data through the Router Agent to cleanly hand off validated intelligence to the Taskmaster Agent—required meticulous engineering to prevent data loss and context drifting.

Accomplishments that we're proud of

We successfully implemented an "Organic Discovery" feature. If the agent finds a new real-world constraint via Google Search (e.g., a specific ferry line is closed in winter), it autonomously writes this rule to a pending_rules database for the Admin to approve. We are also incredibly proud of bridging the gap between a chat interface and real-world execution by enabling our agent to autonomously send structured emails and calendar invites.

What we learned

We learned the immense power of combining structured RAG (Firestore) with real-time Google Search Grounding to eliminate AI hallucinations. We also discovered that separating concerns into a Multi-Agent system (Router vs. Taskmaster) is vastly superior to relying on a single omnipotent agent.

What's next for Xolvia OS

Now that Voice UI and real-time streaming are fully operational, our next phase includes:

  • Real-World API Integrations: Connecting with platforms like Booking.com or Skyscanner to allow Xolvia to securely and autonomously purchase tickets and book hotels upon user approval.
  • IoT Smart Home Hookups: Integrating with smart home ecosystems so Xolvia can prepare your house (adjusting AC, turning on lights) as you return from your orchestrated trip.
  • Native Mobile Experience: Evolving the platform into a React Native application for a truly on-the-go autonomous concierge.
Share this project:

Updates