# Data Rabbit — your database has a desktop now

This is a markdown copy of [https://rvbbit.ai/data-rabbit/](https://rvbbit.ai/data-rabbit/). The deep reference is [https://rvbbit.ai/docs/data-rabbit/](https://rvbbit.ai/docs/data-rabbit/).

Data Rabbit is a spatial SQL desktop for Postgres: live schemas, reactive blocks, dashboards, and an assistant with receipts. Useful on any Postgres. Alive on RVBBIT.

## Your database has a desktop now

Stop visiting your data through little browser tabs. Data Rabbit turns Postgres into a place you can enter—open windows, follow ideas, arrange evidence, and leave the whole investigation waiting for tomorrow.

![The Data Rabbit desktop with Finder, SQL, data search, and adaptive routing windows](https://rvbbit.ai/data-rabbit/shots/data-rabbit-desktop.png)

## 01 / The territory — a spatial memory for analytical work

### Not tabs. Territory.

A query is rarely the whole thought. The table you inspected, the chart that surprised you, the second query that explained it—those belong together. Data Rabbit lets the shape of the work survive.

![SQL and chart windows arranged across the Data Rabbit desktop](https://rvbbit.ai/data-rabbit/shots/desktop-canvas.png)

![A Finder for Postgres schemas with live table information](https://rvbbit.ai/data-rabbit/shots/finder.png)

One canvas. Many lines of thought. Save all of it as a Scene.

1. **Find.** Schemas become a living Finder: rows, heat, freshness, storage and history.
2. **Think.** Keep SQL, results, charts and context arranged together on an infinite canvas.
3. **Make.** Turn reactive queries into a dashboard without changing tools—or hiding the SQL.
4. **Remember.** Save the entire investigation as a Scene. Reopen the thought, not just the file.

![The Scenes browser: saved desktops as live screenshots with derived facet chips and search](https://rvbbit.ai/data-rabbit/shots/scenes-browser.jpg)

Every saved desktop, photographed. Search by table, window, or connection — one click restores the entire investigation.

## 02 / The motion — a thought can become an app

### Query → result → chart → artifact.

SQL blocks reference each other with `{block}` and re-run when upstream results change. Lineage stays visible. Drop a result on the chart shelf, compose a dashboard, publish it. No export ritual in the middle.

1. **Connect — Reactive SQL.** A spreadsheet’s recalculation model, except every cell is readable SQL.
2. **Shape — Chart shelf.** Arrange dimensions and measures directly over the real result set.
3. **Ship — SQL-authored app.** A live URL for the team. Underneath, SQL you can still inspect and version.

![Reactive SQL blocks connected by lineage](https://rvbbit.ai/data-rabbit/shots/reactive-blocks.png)

![Chart controls over a live query result](https://rvbbit.ai/data-rabbit/shots/chart-shelf.png)

![A dashboard authored from SQL inside Data Rabbit](https://rvbbit.ai/data-rabbit/shots/sql-dashboard.png)

## 03 / The intelligence — an assistant with evidence

### Ask the warehouse. Get receipts.

Search data by what it means. Ask your configured assistant to investigate, run governed SQL, build blocks, or make a dashboard. Every model-backed result names the model version that produced it. Trust is a column, not a vibe.

- Speaks MCP both ways
- Remembers what the team built
- Shows actions, outputs, and receipts

![Semantic data search ranking database objects by meaning](https://rvbbit.ai/data-rabbit/shots/data-search.png)

![A visual semantic operator canvas with model configuration and output](https://rvbbit.ai/data-rabbit/shots/operator-canvas.png)

## 04 / The language — semantic operators in ordinary SQL

### A WHERE clause that can read.

Not a chatbot bolted onto a database — new verbs inside the language you already speak. Judgment and meaning as plain SQL functions, every answer receipted with the exact model that produced it.

```sql
SELECT title, county,
  rvbbit.means(observed, 'aggressive encounter') AS aggressive,
  rvbbit.about(observed, 'nighttime') AS night_score
FROM bigfoot.sightings
WHERE rvbbit.means(observed, 'a credible first-hand report')
ORDER BY night_score DESC;
```

An operator that builds operators — born with its own tests:

```sql
SELECT rvbbit.clover_llm_make_operator(
  'TRUE if a review mentions a competitor');
```

## 05 / The timeline — table time travel, made physical

### Drag yesterday.

Accelerated tables keep their history as scrubbable generations. Pull the handle and watch the table “as of” any moment — diff a metric against last night, explain the change, settle the argument before coffee.

![Dragging the time-travel scrubber across a table's generation history](https://rvbbit.ai/data-rabbit/shots/time-travel-scrubber.png)

## 06 / Below the surface — the warehouse has depth

### Boring Postgres. Unreasonable superpowers.

Put [RVBBIT](https://rvbbit.ai) underneath and the same desktop becomes a semantic, time-traveling, self-accelerating warehouse. One extension. No hosted silo. Your database stays yours.

1. **Scry.** Follow the catalog’s hidden paths. Spider from any table through everything known about it.
2. **Capability packs.** Install new senses. Tested operators, versioned models, one click.
3. **Adaptive routing.** Every query finds its engine. Native / DataFusion / Duck—chosen and explained.
4. **Model Studio.** Make prediction a SQL verb. `train_model()` → `predict_*()`

![Scry visualizing relationships around a database table](https://rvbbit.ai/data-rabbit/shots/scry.png)

![The capability catalog — packs of tested semantic operators ready to install](https://rvbbit.ai/data-rabbit/shots/capabilities.png)

![Adaptive query routing among Postgres, DataFusion, and DuckDB](https://rvbbit.ai/data-rabbit/shots/routing-cockpit.png)

![Model Studio training a machine learning model from SQL](https://rvbbit.ai/data-rabbit/shots/model-studio.png)

## 07 / Field transmissions — what happens when the pieces cooperate

### Moments a feature list cannot explain.

> “Someone on your team built something similar yesterday. Want to start from theirs?”

The assistant offered a colleague’s previous work while an analyst was still sketching. Nobody programmed that moment.

> `WHERE rvbbit.means(observed, 'sounds like a hoax')`

One clause read five thousand reports and took a position on each—batched on GPUs, cached, receipted.

> “The metric moved overnight. I diffed the table against yesterday before coffee.”

Time travel is not only a backup feature. Sometimes it is an argument-settling feature.

> “We asked her what the business should build next. She read fifty conversations and handed us a project plan.”

Two weeks of people simply asking questions became a synthesis of what they were all actually trying to decide. The stakeholders nearly fell over.

## 08 / The layers — honest about what runs where

### Useful on Postgres. Alive on RVBBIT.

**Base layer, any Postgres**

- Desktop windows, Scenes, spatial canvas
- Finder with live schema browsing
- SQL editor, reactive blocks, lineage
- Chart shelf and SQL-authored dashboards
- CSV import with schema inference

**Power layer, RVBBIT underneath**

- Semantic operators in ordinary SQL
- Table time travel and visual diffing
- Columnar acceleration + adaptive routing
- Versioned capability packs and model receipts
- Metrics, cubes, alerts, models, warehouse MCP

## 09 / Entry point — enter your database

Read the field manual for how Data Rabbit runs today on the Postgres you already have. Local state, no hosted detour, every time.

[Read the field manual](https://rvbbit.ai/docs/data-rabbit/)

## Links

- [Docs](https://rvbbit.ai/docs)
- [Benchmarks](https://rvbbit.ai/benchmarks)
- [Capabilities](https://rvbbit.ai/capabilities)
- [RVBBIT](https://rvbbit.ai)
