Pocket Nexus AI Dashboard - v1.2

1 · Upload business data

CSV or Excel. Data stays in your browser except what you explicitly send to the AI gateway below.

2 · Your context (persists across regenerations)

Tell the AI what this data is, and what you're trying to see — it'll shape which KPIs, charts, and framing it picks. Pick a starting point below, or just type your own.

Applies to AI-written insights, commentary, and titles below. Menus stay in English; your data is never translated.

Ask your data

Ask a question in plain language. By default only the data profile (stats, not raw rows) is sent automatically.

AI gateway configuration

Sensitive-looking columns and values are always masked before anything leaves the browser, regardless of these settings.

Response language

Applies to AI-written insights, commentary, chart/report titles, and Ask AI answers. Menus, buttons, and navigation stay in English. Your data (column names, category labels, values) is never translated.

Data sending limits

Getting started

1. Upload data. Drop a CSV or Excel file. Columns that look sensitive (names, emails, phone numbers, IDs, addresses, etc.) are automatically detected and masked — both by column name and by scanning actual values.
2. Data overview appears instantly. A set of charts is computed directly from your real data — no AI call, no cost, always accurate. Click any bar or slice to drill down; use the filter menu to slice by a specific field.
3. Configure the AI gateway (Settings tab). Leave the endpoint as "default" to use the built-in gateway, or type a different one. Add an API key if your gateway needs one. Adjust the timeout if generation is slow.
4. Connection is checked automatically. The first time you generate a dashboard or ask a question, the app quietly confirms the gateway is reachable before proceeding — you don't need to test it yourself first. The dot in the header (top right, clickable — jumps to Settings) shows the current status; use "Test AI Connection" in Settings if you want to check manually or troubleshoot.
5. Describe your context (optional but recommended). Tell the AI what the data is and what you're trying to see — e.g. "This is stock trading data, I'm reviewing as a Sales Director." This context is remembered across regenerations.
6. Generate the dashboard. The AI decides which KPIs/charts are relevant (it never sees your full dataset — it just designs the layout, and your browser computes the real numbers from all your rows). You can regenerate with a new instruction any time; the context above stays, only the specific ask changes.
7. Filter and drill down. Both the data overview and the AI dashboard share the same filters. Clicking a chart segment filters instantly — no AI call. Optionally tick "refresh AI insights on filter change" if you also want the narrative text to update for that slice.
8. Ask your data. Free-form questions go to the AI along with a data profile (stats only, by default). Tick "include full dataset" for questions that need to see more rows, or "keep conversation context" to avoid resending data on follow-ups.

Privacy & data handling

• Sensitive-looking columns and values are masked before anything leaves your browser, regardless of any other setting.
• Dashboard generation only ever sends a statistical profile and a 10-row sample — never your full data — since the AI only designs the chart layout, not the numbers.
• "Ask your data" can send more (up to 5,000 rows / 1MB, masked) but only if you explicitly enable full-dataset mode and confirm consent.
• Nothing is stored on a server by this page — everything runs in your browser, and data resets when you upload a new file or close the tab.

About Nexus Gateway AI

This app talks to an AI gateway to generate dashboards and answer questions. A default gateway is pre-configured, but you can connect your own.

To get your own API key and run this dashboard against your own gateway, visit ai-intel-nexus.ai.studio, then paste the endpoint and key into the Settings tab.

Troubleshooting

• Generation times out: run the latency report next to "Test AI Connection" first — it tells you whether the issue is basic connectivity or the gateway being slow to produce a larger response. Try "Try lighter version" on a failed generation, or raise the timeout.
• Filter menu looks empty or missing a field: filters are built from your data's categorical/date columns automatically — very high-cardinality free-text columns (like unique IDs) may not qualify.
• Copy button fails: the app falls back to a legacy copy method automatically; if both fail, a manual copy box appears with the text pre-selected.
• HTML preview doesn't show: the "Preview HTML" option only appears when a response looks like it contains HTML markup, and renders in a fully sandboxed iframe (no scripts run, by design).

Nexus AI Intelligence Platform

End User License Agreement (EULA), Terms of Service (ToS), and Community License v1.0

Product: Pocket Nexus AI Dashboard  ·  Platform: Nexus AI Intelligence Platform  ·  Version: 1.0  ·  Effective Date: August 19, 2026

Too much data, too little time?
No Excel skills or BI tools? No problem.
Meet your portable AI Business Analyst.
Upload your data and let AI discover insights, trends, and opportunities.
Turn your information into decisions — faster.

About

Pocket Nexus AI Dashboard — part of the Nexus AI Intelligence Platform.

© 2026 Nexus AI Intelligence Platform
Licensed under the Nexus AI Intelligence Platform Community License v1.0 — see the EULA & ToS tab for full terms.

v1.2 — Current

HTML/report responses in "Ask your data" no longer dump raw code into the chat.

• When an AI response is HTML (a report, a formatted answer), the chat bubble no longer shows the raw markup by default — it shows a short "Done — press 'Preview HTML' or 'Open in new window' to view it" message instead. Plain-text answers are unaffected.
• Copy and Save as still work on the full, real response either way — hiding the raw source from the bubble only changes what's displayed, not what's stored.
• The preview toggle button now reads "Preview HTML" / "Hide preview" instead of "Preview HTML" / "Show text," since there's no longer a raw-text view to switch to for HTML responses.
• Fixed: selecting "Presentation 16:9" was still producing dense, A4-style paragraph content — the layout instruction had a label but no actual writing-style guidance, so the AI defaulted to normal document prose regardless of which layout was picked. Both A4 and 16:9 now include a real behavioral cue (paragraph-style vs. short slide-style headings and bullets).
• Fixed: charts could render distorted or broken (overlapping pie wedges, label text buried behind the chart) — SVG-based charts (pie, donut, any hand-computed geometry) are no longer used at all. "Include charts" now only produces simple CSS bar charts, where the AI only has to set a percentage width — nothing it can get visually wrong.
• Architecture change: the AI no longer draws charts for "Ask your data" at all — it only supplies numbers (a small placeholder with labels/values), and the app renders the actual chart (bar via CSS width, pie via a CSS conic-gradient) the same reliable way "Generate Dashboard - AI" already worked. Chart geometry is now always correct by construction, matching the approach that already made Generate Dashboard's charts reliable.
• Added a response language toggle (English, Indonesian, Chinese, Japanese, Korean, Spanish, French, German) — applies to AI-written insights, commentary, titles, and Ask AI answers everywhere; menus/navigation stay English and your actual data (column names, values) is never translated. A dropdown with a short description sits right next to "Generate Dashboard - AI," with a matching one in Settings — both stay in sync.
• Connection is now tested automatically and silently right when the app loads, so the header status dot is accurate before you've clicked anything.

v1.1

Settings, Help, EULA & ToS, About, and gateway metadata added; larger data cap; chat UX refinements.

• Full-dataset send cap for "Ask your data" raised to 5,000 rows / 1MB (from 2,000 rows / 500KB), enforced client-side before anything is sent.
• AI gateway configuration moved out of the main Dashboard flow into its own Settings tab.
• New Help tab additions: a reminder to test the connection and confirm the green status dot before generating a dashboard or asking a question, plus an "About Nexus Gateway AI" section with a link to get your own API key.
• New EULA & ToS tab with the full Nexus AI Intelligence Platform Community License v1.0.
• New About tab with creator attribution and license reference — kept in its own tab so it stays out of the way of normal use.
• Chat message actions (Copy, Preview HTML) moved from the top to the bottom of each AI response.
• HTML preview enlarged (520px, vertically resizable) and can now open full-size in a new browser tab/window via a sandboxed blob URL.
• Fixed: connection could fail on some gateways because of extra request headers triggering strict CORS policies — the gateway request mechanism has been fully reverted to the last known-working version to restore reliable connectivity.
• Added a "Save as" button on AI chat responses — downloads as a standalone .html file for HTML-formatted answers, or .txt otherwise.
• Added a Copy button on your own questions in "Ask your data," so a failed or timed-out question can be resent without retyping.
• Moved the "runs entirely in your browser" note under the header subtitle; footer now shows a subtle, single-line creator credit instead.
• "Data Overview" relabeled to "Data Overview - Local (Non-AI)" to make clear those charts never touch the AI gateway.
• Added guidance and examples for when to check "Include full dataset" on a question in Ask AI.
• "Open in new window" and "Save as" now include Print and Save as buttons (via the browser's native print dialog).
• Connection is now checked automatically before generating a dashboard or asking a question — no more required trip to Settings first. A failed check shows a clear inline message with a one-click link to Settings.
• "Include full dataset" consent for Ask AI moved from Settings to an inline confirmation right on the question itself — no back-and-forth between tabs.
• Header connection status is now clickable, jumping straight to Settings.
• Renamed for clarity: "Test connection" → "Test AI Connection"; "Generate/Regenerate dashboard" → "Generate/Regenerate Dashboard - AI".
• Default AI model changed to "pocketnexusdb".
• Added persona quick-select chips (Sales, Business Owner, Finance, Marketing, Operations, HR, Support, Retail) in "Your Context" — one click fills a starting sentence, still editable.
• "Your Context" now also shapes "Ask your data" answers, not just dashboard generation.
• Added Report options to "Ask your data": output format (text/HTML), layout (none/A4/16:9 presentation), with or without charts (safe inline SVG/CSS, no scripts), and number of pages (1–10 or AI decides). Multi-page HTML answers get automatic Prev/Next navigation in the chat, the new-window view, and the saved file.
• Fixed a critical bug: the multi-page report generator's embedded closing tag (</script>) was written unescaped inside the app's own script block, which caused browsers to truncate the entire app's JavaScript at that point — breaking most functionality. Escaped correctly; verified the app's script now parses in full and generated reports still render correctly.
• Fixed: "Save as" in the new-window/report view was just aliasing to Print — it now genuinely downloads the HTML file, with Print kept as a separate, distinct button.
• Chart quality in HTML reports improved: the AI is now guided to vary chart types (SVG pie/donut for composition, line/area for trends, bar for comparisons) with labeled values and a consistent color palette, and — when charts are enabled — to present the answer as a compact mini-dashboard (KPI cards plus charts) rather than plain text with a single flat bar.
• Removed the "Save as" button from the "Open in new window" view — downloading a file from inside a window opened via a blob URL proved unreliable across browsers. The "Save as" button on the chat response itself (which saves from the app's normal page context) remains the reliable way to save a response as a file; "Print" (including "Save as PDF" from the print dialog) still works in the new-window view.
• Fixed: Report options (layout, charts, page count) were being silently discarded unless Output format was also manually set to HTML — those choices now imply HTML output automatically.
• Rewrote the report-generation instruction sent to the AI to read as a natural request (matching how a person would phrase it when typing directly) instead of a rigid technical spec — this matches the reliability of manually-typed formatting requests. Chart guidance softened from a strict SVG mandate to a simpler "keep charts basic, prefer a table when a chart isn't essential" note, since forcing elaborate SVG geometry was a common source of broken-looking charts.

v1.0

Consolidated feature set at initial release.

• Spec-based AI dashboard generation — the AI designs chart layout only; the browser computes real numbers from your full dataset, keeping AI responses small and fast.
• Data Overview: automatic, AI-free charts grounded in your real data, including profit/loss-style splits, category totals, distributions, and time trends.
• Shared filter menu and click-to-drill-down across both the Data Overview and AI-generated dashboard, updating instantly with no AI call.
• Persistent context field plus per-regeneration instructions, with short-term memory of recent requests.
• Executive summary line computed for every chart, from real values — accurate by construction, no extra AI call.
• Sensitive-data masking by column name and by value pattern (emails, phone numbers, etc.), with an explicit consent flow for sending a larger sample.
• Dual gateway support: default + custom endpoint, optional API key, configurable timeout.
• Connection test with round-trip latency, plus a 3-ping latency report to help distinguish network issues from generation-size issues.
• "Ask your data" chat with session memory, per-message latency, copy-to-clipboard (with fallbacks), and sandboxed HTML preview for responses containing markup.
• CSV export per chart and full summary; print-to-PDF for the dashboard.

Build history

Earlier iterations that led to v1.0.

• Initial rebuild: hardened error handling, XSS-safe rendering, sensitive-value scanning, consent-gated full-dataset mode, CSV/print export, session chat.
• Multiple interactive charts, per-filter re-analyze toggle, 2,000-row/500KB send cap, gateway endpoint hidden behind "default".
• Configurable timeout, elapsed-time indicator, "lighter version" fallback, connection latency display.
• Click-to-drill-down and filter chips added to the Data Overview section.
• Architecture change: AI switched from returning computed chart data to returning a compact layout spec, computed locally — fixed generation timeouts and improved accuracy.
• Filter menu broadened to include high-cardinality columns; shared filtering unified across both dashboards.
• Persistent context + per-regeneration instructions with short-term memory; executive summaries per chart.
• Clipboard copy hardened with legacy and manual fallbacks; sandboxed HTML preview added for AI chat responses.