Data Rabbit

Data Rabbit has a site of its own now — screenshots, field notes, and the quickstart live at datarabbit.com. This page stays the deep reference.

Data Rabbit is RVBBIT's desktop: a fast, local-first SQL client for Postgres built around a windowed canvas instead of tabs. It works against any Postgres - browse, query, chart, monitor - and when it detects the pg_rvbbit extension on a connection, the whole rvbbit surface lights up: operator studios, receipts and costs, adaptive routing, the knowledge graph, cubes, metrics, capability deployment, and more.

Get it#

Three ways to run the same app:

• Native macOS app (Apple silicon) — signed and notarized, no Docker required:

  https://rvbbit.ai/dist/mac/DataRabbit-4.1.6-arm64.dmg
  sha256: f5926145edb52b9648747e96cd117ef2b009c0d6850a897e9c1035afbf081552
  

  The .app bundles its own Node runtime and serves the desktop on a stable private origin (127.0.0.1:43129), so scenes, saved views, and assistant history survive relaunches. Connections and homebase state live in ~/Library/Application Support/Data Rabbit/. All you need is a reachable Postgres; Docker is used only if you install capabilities that launch containers. Requirements and architecture notes live in the repo's docs/MACOS.md.

• Native Linux app (x86_64) — AppImage (chmod +x and run) or .deb:

  https://rvbbit.ai/dist/linux/DataRabbit-4.1.6-x86_64.AppImage
  https://rvbbit.ai/dist/linux/DataRabbit-4.1.6-amd64.deb
  

  Checksums sit beside each artifact (.sha256). Same architecture as the Mac app — bundled runtime, stable private origin, state under ~/.config/Data Rabbit/.

• The Docker ensemble — ships alongside the warehouse and serves on port 3000; "install" is opening http://localhost:3000.

• Bare web app — it is a Next.js app at heart; the container image is published as ghcr.io/ryrobes/rvbbit-lens (the project's earlier name; the image name follows the rename later).

The macOS shell keeps Node fully isolated from the renderer and drives the same product actions as the in-canvas menu bar, so browser and .app behavior cannot drift.

Connections#

Connections are managed in the Connections window (Database menu → Connections…). Data Rabbit connects server-side through a pooled pg driver, so credentials never reach the browser; they are stored in a config file on the server (connections.json under the app's data directory - a Docker volume in the ensemble, ~/.config/rvbbit-lens/ on a bare install) with passwords redacted from every API response.

• Discrete fields (host, port, database, user, password, SSL mode) or a full postgres:// connection string.
• Per-connection SSH tunnel support (bastion host, private key or password) for reaching databases behind a jump box.
• A Test connection action that reports schema/table counts before you save.
• The connection picker in the menu bar shows label · database; an æ badge marks rvbbit-enabled connections.

Two pool lanes keep the UI responsive: interactive queries get the main pool while monitors and schema fan-out use a small isolated meta pool, so a runaway dashboard can't starve your query.

Hostnames are resolved by the Data Rabbit server, not your browser. Because connections are made server-side, the host you type is answered from wherever the lens process runs. In the Docker ensemble that means localhost points at the lens container itself - use the compose service name instead: host postgres, port 5432 (not the host-published 55433). Conversely, if Data Rabbit runs directly on your machine and you're reaching a remote database through an SSH tunnel, then localhost:<tunnel-port> is exactly right - the tunnel lives where the server lives.

The Desktop#

Windows live on an infinite pan/zoom canvas. Each window is draggable, resizable, and minimizable, and the desktop itself is a first-class object:

• Workspaces - five scratch workspaces plus a Scene slot, switched with Alt+1…5 and Alt+6. Occupancy dots in the menu bar show which are in use.
• Scenes - named, saveable desktops (Save / Save As / Open / Rename / Delete from the Scene tray). Scenes can be shared and forked between "homes".
• Persistence - the desktop persists locally first (browser storage) with a server-side shadow (SQLite), so state survives restarts and can follow a shareable home URL.
• Present mode - a read-only toggle that strips window chrome and editor rails and fits the canvas to the screen, for demos and wall dashboards.
• Theming - light/dark themes, a full Palette editor (Desktop menu → Palette…) over the OKLCH design tokens, and wallpaper support that can derive an accent palette from the image - including an optional vision-model path that themes the whole desktop from your wallpaper.
• Fonts - Desktop menu → Font: sans family, mono family, and UI size.
• Dependency lines - an optional overlay that draws curves between SQL windows that reference each other, so the reactive graph is visible.

SQL Windows#

⌘N (File → New SQL window) opens the core surface: a CodeMirror SQL editor with completion and formatting over a virtualized result grid.

• Run with ⌘↩; transaction controls (BEGIN / COMMIT / ROLLBACK); per-window query history; open a .sql file from disk.
• Body tabs: rows (grid), profile (column profiling), chart, sql, explain (visual plan graph), steps (semantic-operator step trace after a Cascade runs), and app (SQL-authored dashboard artifacts).
• Export: CSV, JSON, copy as INSERTs, plus "Save as view" and "Save as canonical viz block" on rvbbit connections. Large results page in with "Fetch 5000 more rows".
• Ask mode - type plain English; the window generates SQL through rvbbit.synth_sql and shows the query before running it.
• HTML Block mode - chat-author an HTML app backed by named SQL queries (rvbbit.html_block_turn), rendered live in the app tab.

Reactive Blocks And Params#

Every SQL window has a block name, and any other window can reference it as {block_name} - rewritten at run time as an inline subquery. Clicking a cell in a grid (or a mark in a chart) emits a filter param onto the workspace; windows subscribed to that param re-run automatically, and filtering an upstream block cascades into every downstream reference. Block chips are draggable: drop one on the canvas to spawn SELECT * FROM {block}, or onto a rollup tile to pipeline it.

This is the desktop's quiet superpower: a handful of small SQL windows compose into a live, cross-filtered dashboard without a dashboard builder.

Charts#

The chart tab auto-infers a Vega-Lite spec from your result columns, themed from the active palette. Two editors sit on top:

• Chart shelf - a Tableau-style drag-and-drop shelf editor (fields onto x/y/color/size shelves, aggregate and time-unit pickers) that round-trips with the underlying Vega-Lite spec; an "Edit spec" YAML pane is always available for full control.
• Rollup shelf - drag column chips onto group-by / measure / pivot / top-N tiles to generate aggregation SQL, including semantic-operator measures on rvbbit connections.

Chart marks participate in the param system - clicking a bar filters downstream windows, same as clicking a grid cell.

Time Travel Scrubber#

On an rvbbit connection, tables registered for acceleration get a time-travel scrubber: a strata-style timeline of the table's generations (built from rvbbit.time_travel_timeline, so it never scans data). Drag the handle to pin the window to an AS OF moment - the editor is rewritten with the -- rvbbit: as_of = '…' comment and re-run - hover ticks for snapshot cards, or type an exact datetime. See Time Travel.

Finding Things#

• Command palette (⌘P) - fuzzy quick-open over tables, saved views, windows, and tools.
• Scry (⌘K) - a full-screen semantic search canvas. Each search stage scopes within the previous stage's results ("tables about shipping … within those, columns about delay"), and editing an upstream stage re-flows everything downstream. Results spawn into data windows.
• Finder (⌘F) - a schema tree grouped by namespace with search, DDL view, and inline table-activity sparklines.
• Data Search - free-text semantic search over the catalog, ranking tables and columns by what their data is about.

The RVBBIT Cockpits#

On an rvbbit connection the Database menu grows a full set of windows, each a thin, live view over the SQL surfaces documented elsewhere in these docs:

Window	Over	Docs
Receipts / Costs / Cache	rvbbit.receipts, cost ledgers, embedding/result caches	Receipts And Costs
Operator Studio	rvbbit.operators - author, inspect, and try operators on a canvas, with step graphs and receipt timelines	Semantic SQL, Cascades
Specialists / Model Studio	model backends, training runs, evaluations	Providers, Predictive Models
Adaptive Routing	route decisions, training, profiles, workload layouts, freshness	Routing And Training
MCP Servers / MCP Incoming	server registry, tools, invocations; incoming agent traffic	MCP Servers
Capabilities	the capability catalog, install graphs, and a Hugging Face deploy flow	Capability Packs
Warren	nodes, jobs, deployments	Warren
Duck Monitor	Duck/Vortex worker telemetry	Duck/Vortex Worker
Knowledge Graph (browser, graph explorer, extraction runs, merge review)	rvbbit.kg_*, with a force-directed graph canvas	Knowledge Graph
Metrics (catalog, creator, inspector, board)	metric definitions, bitemporal history	Metrics And KPIs
Cube Studio (catalog, creator, inspector, proposals)	cube definitions, health, enrichment	Cubes
Alerts	alert rules, state, queue, firing history	Alerts
Drift	catalog drift detection	Catalog
Query Lens	per-query drill-down: receipts, lineage, cost	Receipts And Costs
Scheduler tray	pg_cron jobs, with one-click presets for the maintenance entrypoints	Accelerator Freshness

The Metric Inspector is a nice example of the house style: a definition-time version picker on one axis and a data-time time-travel scrubber on the other, so "what did this KPI say, under which rule, over which data" is one window.

Postgres Tools#

The non-rvbbit toolset works on any connection:

• Postgres Monitor - live activity (polling stands down when its workspace is parked).
• Postgres Admin - object management (roles, grants, tables).
• System Objects and Extensions browsers.
• Notification Center - LISTEN/NOTIFY streamed to toasts and a feed.
• CSV Import - streaming multi-gigabyte CSV import with dialect and date format inference and generated DDL.
• View Apps - save any SQL statement as a launchable desktop "app" with its own icon.

SQL-Authored Dashboards#

Results can be UI: a statement that returns rvbbit_artifact = 'ui' rows renders as dashboard tiles - charts, metric cards, KPI gauges, sparklines, tables, filter controls, action buttons - laid out by weight. Filter controls bind back into the same param system as everything else. Viz Blocks makes these reusable: versioned SQL templates stored in rvbbit.viz_block_defs (rvbbit.define_viz_block, rvbbit.preview_viz_block) that can be linked to tables and shipped with the database rather than the client. Published dashboards render in a sandboxed Dashboards window with a read-only query bridge.

Keyboard Shortcuts#

Shortcut	Action
⌘N / Ctrl+N	New SQL window
⌘↩	Run the focused SQL editor
⌘P / Ctrl+P	Command palette
⌘K / Ctrl+K	Scry semantic search canvas
⌘F / Ctrl+F	Finder (schema browser)
⌘Z / ⇧⌘Z	Undo / redo desktop actions
Alt+1…5, Alt+6	Switch workspace / Scene slot