Free to use — bring your own AI keys, pay providers directly

Dashboards that track your situation and tell you what changed

Connect your data and pin the questions that matter. One click keeps both your numbers and the live web in sync — and shows you exactly what moved since last time. Your data and the world, on one screen, always current.

The demo is a real dashboard — no sign-up, no AI key. Opens instantly.
Live data + web research·What-changed on every update·Multi-model AI council·Postgres / MySQL / SQLite / REST
Revenue board · liveupdated just now
Total revenue
$3.43M
▲ 12.4%
What changed
Revenue3.43M → 3.85M▲ 12.4%
3 new sourceson “pricing pressure”
Consensusmedium → high

A dashboard tells you what happened. Nothing tells you what changed.

So you build the dashboard once, then re-check it by hand — reload the data, re-run the same searches, try to remember last week’s numbers. The moment that actually matters — the number that moved, the source that appeared — is the one you find out about late.

datavisual.studio keeps your data and the live web in sync behind one dashboard, and leads with the delta: what moved, by how much, and what’s new — since the last time you looked.

Since your last visit
Revenue3.43M → 3.85M ▲ 12.4%
Churn2.1% → 2.6% ▲ 0.5pt
Pinned research3 new sources
Illustrative — this is what an Update surfaces.

Everything between raw data and a decision

The dashboard work you’d do in a BI suite and the research you’d do in five browser tabs — in one place, and alive.

The living monitor

One update, and it tells you what changed

Pin the metrics and research questions you care about. Hit Update and the dashboard re-pulls your live data and re-runs every pinned topic — then hands you a plain-English feed of what moved, by how much, and which new sources appeared.
Instant dashboards

A full dashboard from your data in seconds

Metrics, interactive charts, entity comparison and a searchable data table — generated from your columns in one click, no AI cost. Five templates and a focus column shape the first draft; every widget is fully rebuildable.
Edit by chat or by hand

“Add a pie of revenue by product.” Done.

Talk to the dashboard assistant, or use the component gallery, form editors and reorder controls. Every edit updates the same dashboard in place — it understands what you mean, typos and all.
Deep research, cited

A council of AI models, not one confident guess

One model can be confidently wrong. Here, several models answer independently, review each other anonymously, and a chairman synthesises one cited report — grounded in live web research before it ever answers. The findings land straight on a dashboard.

And everything around it

Connect your database

PostgreSQL, MySQL, SQLite or any JSON REST API — read-only, so the dashboard reflects the source, not a stale export.

Share as a link

Turn any dashboard or report into a public, read-only link. No sign-in for viewers; your data source and keys stay private.

Exports that hold up

Structured PDF or self-contained HTML — charts embedded as images, rendered markdown and cited sources.

Templates & focus

Minimal, Overview, Full, KPI or Visual — then pick the column the headline metric and first charts revolve around.

Live data table

Every dataset comes with a searchable, sortable, paginated table and one-click CSV download.

Prediction engine

When your data carries ratings, a deterministic ELO / Poisson / XGBoost engine adds calibrated forecasts.

Built for anyone who asks their data questions

One tool that flexes from a quick chart to a fully-monitored, cited research board.

Analysts

Skip the BI-tool setup — a full dashboard from a CSV or query in seconds, then refine it in plain English.

Founders & operators

Watch the metrics that matter and the market around them; one Update tells you what moved since yesterday.

Researchers

Ask a hard question and get a cited, multi-model report grounded in live web research — not one model’s guess.

Teams

Share a live, read-only link so anyone can see the dashboard — no seats, no sign-in, no exposed keys.

Two workflows, one living artifact

Build a dashboardno AI cost
1
Bring data

Upload CSV / Excel / JSON, or connect a database with a read-only query.

2
Generate

Pick a template — metrics, charts, comparison and table appear instantly, every widget rebuildable.

3
Refine

One-click components, form editors, or just tell the assistant what you want.

4
Monitor

Hit Update to re-pull the source and see what changed — your edits survive.

Deep research~2 min
1
Ask anything

With or without a dataset attached. Watch every step live.

2
Web + council

Targeted web searches gather current facts; multiple models analyse and peer-review anonymously.

3
Cited report

A chairman synthesises one calibrated answer — sources, consensus and prediction tables.

4
Becomes a dashboard

The findings, analytics and charts land on a live dashboard you keep editing and monitoring.

How it’s built

This one’s for the engineers. The interesting part of this product isn’t the feature list — it’s the constraints it holds so the numbers can be trusted. A stranger’s browser talks to a thin Next.js proxy, which is the only thing that reaches the FastAPI backend; all state is JSON on disk.

BrowserA stranger's device
Next.js proxyVercel · verifies the session
FastAPIAWS · the engine
data/ · JSONlocal disk · the whole DB
Request path: the browser talks only to a Next.js proxy on Vercel, which verifies the Clerk session and forwards to a FastAPI backend on AWS, which keeps all state as JSON files on local disk — there is no separate database.

Next.js 16 · React 19 on Vercel · FastAPI · pandas · Plotly on AWS · Clerk for identity · JSON on disk. No database.

The model never does arithmetic

The LLM turns your question into a JSON query spec; a deterministic Python engine executes it and phrases the answer from the result. Nothing about a number is left to the model — so every figure the product shows can be defended, because a person or a test can re-derive it.

Numbers fail closed

Every figure in an answer must appear in the computed result or be a derivation the answer shows its working for — otherwise it refuses rather than guess. A level like MRR is never summed across time; “total customers in June” aggregates across plans, not one of them.

Your keys, never mine

Users bring their own AI keys, encrypted at rest and resolved per request from the signed-in identity. The owner’s key is never spent on someone else’s request — so the product can be free without a metered bill in the middle.

A council, not one oracle

For research, several models answer independently, review each other’s answers anonymously, and a chairman synthesises one cited report. No single model’s confidence is load-bearing, and a model that quietly drops out is surfaced, not hidden.

No database, on purpose

All state is JSON on disk. For a single-tenant, single-replica product that removes migrations, a connection pool and a second backup target — and makes the security model tractable: one directory is the trust boundary, one function is the id check.

Accessible by construction

Charts carry a text alternative generated from the plotted numbers (not the LLM) and a “view as table” toggle; overlays trap and restore focus; contrast meets WCAG AA. axe runs on every route in CI. The a11y is in the components, not bolted on after.

Pricing, plainly

The app is free and open source. You bring your own AI key (OpenRouter, and optionally Google Gemini) and pay the provider directly for what you run — there’s no markup and no bill from us. Rough real costs, so money is never a surprise:

Build a dashboard
$0

Charts, metrics and the table are computed deterministically — no AI call at all.

Ask a question / edit
~$0.001

One fast-model call to turn your words into a query spec. Fractions of a cent.

A full research run
~2–20¢

3 web searches + a multi-model council. Depends entirely on the models you choose.

No card, and no AI key, to start — the live demo needs neither.

Questions, answered

Is datavisual.studio really free?+

Yes. The app is free — you bring your own AI provider keys (OpenRouter, and optionally Google Gemini) and pay those providers directly for usage. We never bill you, and your keys are only ever sent to the AI providers themselves.

What data can I connect?+

Upload a CSV, Excel or JSON file, or connect a SQL database (PostgreSQL, MySQL, SQLite) or any JSON REST API. Database imports are read-only, and your dashboard can re-pull from the source anytime.

How is this different from a normal BI tool?+

Dashboards are editable by chat and by hand, and they don’t just display your numbers. One “Update” re-pulls your data and re-runs the research questions you pinned, then shows a clear feed of exactly what changed since last time.

What is the AI research council?+

Instead of trusting one model, several AI models answer your question independently, review each other’s answers anonymously, and a chairman model synthesises a single cited report — grounded in live web search.

Can I share a dashboard or report?+

Yes. Any dashboard or research report can be shared as a public, read-only link. Viewers can explore the charts and data but can’t edit anything, and your data source and AI keys are never exposed.

Where is my data stored?+

On the platform’s own server, on local disk — there is no third-party database. Your datasets, dashboards and reports are private to your account and are never sold or shared.

Your data has answers. Go get them.

Free to use — sign in and start building living dashboards in minutes.

Open the studio