AuraQL — Zero-Server AI Analytics Engine

🔮 Inspiration: The Broken Cloud Analytics Paradox

Traditional cloud business intelligence requires piping sensitive customer records, financial logs, and transaction streams to remote third-party databases. When analysts consult AI, language models frequently hallucinate arithmetic calculations on large datasets.

We built AuraQL to completely invert this paradigm:

Zero-Server In-Browser OLAP: The columnar database and vectorized SQL evaluator run 100% inside your client-side browser tab memory.
Deterministic AI Math: The AI agent writes the SQL query, but the in-browser engine evaluates aggregations (SUM, AVG, COUNT, STDDEV) in under 10 milliseconds with mathematical accuracy.
WebMCP Dual-Control Canvas: Both the human user and AI agent collaborate simultaneously on the same dashboard canvas via the emerging open standard document.modelContext.

⚡ Why Our Use Case is a Strong Fit for WebMCP

Analytics dashboards are notoriously difficult for computer-vision and click-based agents: DOM selectors break, dynamic canvas elements cannot be inspected, and SVG splines are opaque to LLMs.

WebMCP provides the ideal interface:

Instead of brittle DOM scraping, AuraQL exposes structured analytical primitives directly to the agent.
The agent calls list_tables_and_schema to discover table schemas and column datatypes.
It evaluates relational SQL via execute_sql_query.
It commands the user's viewport directly via render_interactive_chart and apply_dashboard_filter.

🚀 How It Creates a Better User Experience

100% Client-Side Privacy: Financial data and customer churn metrics never leave the user's browser RAM. Zero cloud data storage costs, zero telemetry tracking, and zero database credentials to expose.
Sub-10ms Latency: Queries against columnar memory buffers execute faster than a network roundtrip to a cloud data warehouse.
Co-Pilot Simultaneous Vision: When you ask the agent a question like "Show quarterly revenue breakdown by region as a bar chart", the dashboard mutates in real time before your eyes while the agent explains the business takeaways.
Pro Studio Interface: Dual-axis resizable grid windows, high-contrast borders (black in light mode, neon purple in dark mode), and drag-and-drop CSV/JSON ingestion.

🤝 What People and Agents Can Do Together That Was Impossible Before

Natural Language Scenario Simulation: Humans specify hypothetical business conditions ("What if critical churn accounts decrease 25% and MRR increases 15%?"). The agent calls simulate_forecast_scenario to execute in-memory adjustments and render baseline vs. projected variance charts immediately.
Instant Statistical Anomaly Isolation: The agent uses detect_anomalies to calculate Z-scores ($|Z| \ge 1.85$) and Interquartile Ranges (IQR) directly in browser memory, isolating revenue spikes or margin compression without needing Python or external Jupyter notebooks.
True Cross-Platform Agent Access: The exact same WebMCP tools are available inside:
- ChatGPT’s native in-app browser (document.modelContext)
- Desktop ChatGPT & Claude Desktop via Server-Sent Events (/sse)
- OpenAI Codex CLI over Stdio JSON-RPC 2.0

🛠️ How We Implemented WebMCP

AuraQL implements the official W3C WebMCP specification:

Native Browser Standard: Inspects document.modelContext on mount and registers 6 analytical tools:
- list_tables_and_schema (Schema & metadata discovery)
- execute_sql_query (Columnar SQL aggregations)
- render_interactive_chart (Viewport visualization mutation)
- apply_dashboard_filter (Cohort isolation)
- simulate_forecast_scenario (What-if financial modeling)
- detect_anomalies (Statistical anomaly detection)
Universal Polyfill & Bridge:
- If running in standard Chrome or without native browser flags, AuraQL binds a spec-compliant fallback to document.modelContext and window.modelContext.
- An external Node.js WebMCP bridge deployed on Render (scripts/mcp-bridge.mjs) synchronizes with the active browser tab via Server-Sent Events (/api/bridge/events) with 25-second keepalive heartbeats, allowing desktop clients (Codex CLI, Claude Desktop, ChatGPT Desktop) to steer the live web app remotely.