Most AI tools upload your entire dataset to their servers. QueryVeil keeps your files in your browser and gives you full control over what the AI can see.
When you drop a CSV, Excel, JSON, or Parquet file, it's loaded into DuckDB WebAssembly — an in-browser SQL engine. The file is read by JavaScript's File API and never sent over the network.
QueryVeil reads your table structure — column names, data types, and nullable flags — plus lightweight column statistics (value ranges, top categories). This metadata helps the AI write accurate queries. No individual rows are included.
Your question and schema metadata are sent to the AI model, which returns a SQL query. How much data the AI sees depends on the mode you choose:
AI sees schema + stats only. Returns one SQL query. Your actual rows never leave the browser.
AI runs multi-step analysis. Capped query results (max 100 rows) are shared so the agent can reason across steps.
The generated SQL query runs against your local DuckDB instance — entirely in your browser. Results are rendered directly in the UI. No data round-trip to any server.
For maximum privacy, enable Zero Cloud Mode. All AI processing runs locally on your device using WebLLM (in-browser) or Ollama (local server). Zero network requests to external AI services.
Zero Cloud Mode
All processing on your device. Verify in DevTools > Network tab.
Open your browser's DevTools and go to the Network tab. In Quick Query mode, you'll see only schema metadata in AI requests — no data rows. In Deep Analysis mode, you can inspect exactly which query results are shared.
Press F12 or Cmd+Opt+I. Navigate to the Network tab.
Drop a CSV or Excel file into QueryVeil.
In Quick Query mode, outbound requests contain only schema and your question — no data rows.
| Privacy aspect | QueryVeil | ChatGPT | Julius AI |
|---|---|---|---|
| Data processed locally | No | No | |
| Files never uploaded | No | No | |
| AI sees schema only (Quick mode) | No | No | |
| You control what AI sees | No | No | |
| Fully offline option | No | No | |
| Verifiable (DevTools) | No | No | |
| Data not used for training | Paid only | Policy | |
| No account required | No | No |
QueryVeil's architecture is designed to be compatible with strict data handling requirements. Since data processing happens entirely in the user's browser, many compliance concerns are addressed by design.
In Quick Query and Zero Cloud modes, no PHI is transmitted to AI services. In Deep Analysis mode, capped query results are shared with the AI provider — avoid this mode with sensitive patient data, or use Zero Cloud Mode for full isolation.
Files are never uploaded to external servers. In Quick Query mode, only schema metadata reaches the AI. In Deep Analysis mode, query results (up to 100 rows) are sent. Zero Cloud Mode keeps all processing on-device.
Your raw files never leave your browser. In Zero Cloud Mode, all processing happens locally — no data crosses borders or enters third-party infrastructure.
QueryVeil is not certified for HIPAA, SOC2, or other compliance frameworks. These architectural properties may support your compliance posture — consult your compliance team.
Try the live demo with sample data. Open DevTools and see exactly what's sent to the AI — nothing more than you'd expect.