What's inside

A quick tour of the platform. Every new account includes a sample dataset and dashboard to explore immediately.

Getting data in

File upload. CSV, Excel (.xlsx), Parquet, JSON — typed, profiled, and previewed automatically on import.

Live connections. PostgreSQL and MySQL with read-only credentials (encrypted at rest), plus Google Sheets / REST JSON feeds on a schedule.

Transform steps. A Power-Query-style applied-steps pipeline: rename, retype, filter, split/merge, calculated columns (SQL), group, unpivot. Raw data is preserved; steps re-run on every refresh.

Refresh. Cron schedules with no daily caps, full run history, and failure alerts to the activity feed.

Modeling

Semantic measures. Named metrics written as plain SQL aggregates (e.g. sum(profit)/sum(revenue)) — the DAX replacement. Reusable in charts and by the AI analyst.

Synonyms & descriptions. Per-column metadata that grounds natural-language questions.

Hierarchies & relationships. Drill-down paths (Year ▸ Quarter ▸ Month) and cross-dataset joins.

Row-level security. Per-dataset rules with {{attribute}} templates bound to each member. Applies to viewers everywhere — including AI queries — and fails closed.

Analyzing

Charts. 17 types from KPI cards to heatmaps, waterfalls, boxplots, and pivots — with stacking, reference lines, and conditional formatting.

Dashboards. Drag-and-resize grids, dashboard filters, cross-filtering by clicking any chart element, drill-through to raw rows, auto-refresh.

SQL editor. Guarded read-only DuckDB SQL with schema autocomplete and CSV/Excel export (editor role).

Alerts. Fixed thresholds or statistical anomaly detection (learned baselines), checked on your schedule, delivered in-app, by email, and to Slack-compatible webhooks.

AI analyst

Ask anything. The agent plans, queries the semantic layer (RLS enforced), renders live charts, and explains — with every executed SQL visible in the panel and logged to the audit trail.

Dashboards on demand. “Build me a sales overview” proposes a full dashboard you approve with one click.

Your model, your choice. Configure Anthropic, any OpenAI-compatible endpoint (custom base URL + key), or local Ollama — per user, in Settings.

Autonomous insights. A nightly scan flags metric anomalies and posts narrated insight cards; every upload gets instant starter questions.

Voice

Push to talk. Hold the mic, ask your question, release — transcribed by self-hosted Whisper and answered by the AI analyst.

Spoken answers. Toggle voice replies to hear the summary (Kokoro TTS, on your server) while charts render.

Commands too. “Add this to my sales dashboard”, “filter to 2026” — voice rides the same governed tools.

Sharing & operations

Sharing. Workspace roles (admin/editor/viewer), public read-only links with optional passwords, iframe embeds, and scheduled PDF email subscriptions.

Export. Pixel-perfect PNG/PDF rendering of any dashboard.

Admin. Org owners see users, AI token spend, and refresh failures at /admin.

Self-hosted. Docker Compose stack: Postgres, Redis, DuckDB-per-workspace, Celery, and a voice container. Nightly backups; MIT/Apache-2.0 dependencies only.