WebKaya

Private AI dashboards from your files, Google Sheets, and web pages. Runs entirely on your device — nothing is ever uploaded.

Ask a question about your data and get a dashboard you can trust — built entirely on your device. WebKaya is a local-first analytics extension: your files are parsed in your browser, the AI runs on your machine, every number is independently re-checked in an isolated sandbox, and a live egress meter proves it ("0 B sent off-device"). ANALYZE ANYTHING YOU ALREADY HAVE • Files: CSV, TSV, Excel workbooks (per-sheet), PDF — parsed locally, with automatic cleanup (placeholder blanks, currency formats, messy headers). • The page you're viewing: one click extracts the visible table, with optional pagination and background collection. • Pasted ranges (great from Google Sheets) and JSON. • Public datasets: import directly from Hugging Face or Kaggle with your own token/key — downloaded straight to your device, persisted locally. ANSWERS YOU CAN VERIFY • Plain-language questions become deterministic computations — sliced, aggregated, and independently re-checked in a WebAssembly sandbox. • The Explore bar surfaces every analysis your data supports: overview, trends, distributions, relationships, causal drivers ("what drives revenue?"), what-if simulation, multi-step comparisons, and stitching two sources together. • Anti-hallucination guards: model-extracted rows must appear in the source document, and any number in the AI's narrative that can't be traced to the verified stats is flagged. • Export everything locally: self-contained HTML reports, totals as CSV, the cleaned dataset, chart data, PNGs. A FULL FINANCE WORKSPACE Switch the workspace selector to Finance for a chart-first terminal over any OHLCV data: candles, volume candles, Renko, Kagi, Line Break, Point & Figure; custom timeframes down to seconds, tick and range bars; indicators on indicators (SMA/EMA/RSI/MACD/Bollinger/VWAP/ATR); custom-formula charts; volume & TPO profiles; volume footprint; bar replay; auto chart patterns; multiple watchlists; multi-condition alerts that never expire — all computed locally and deterministically. YOUR AI, YOUR CHOICE • Built-in model (WebLLM) runs in your browser on your GPU — zero setup. • Or use any model you've downloaded with Ollama — stronger, still 100% local. Reasoning models (DeepSeek-R1 etc.) fully supported. • Optional bring-your-own-key cloud engines (Anthropic / OpenAI): paste your own API key when local models aren't enough. The trust bar turns amber ("Cloud Engine Active") the moment you do, and a local masking pass replaces names/IDs with aliases and masks card numbers, emails and phone numbers before anything is sent. One click switches back to local. PRIVACY IS THE ARCHITECTURE No servers. No accounts. No analytics. No telemetry. The developer receives nothing — ever. A live meter counts every byte that leaves the device, and Strict Privacy Mode blocks all non-local requests outright. Full policy: https://www.webkaya.com/privacy.html