Data Rabbit is a desktop OS for your warehouse — windows, a Finder for your schemas, live dashboards, and an assistant that remembers what your team built. It runs against any Postgres. It comes alive on RVBBIT.
Queries live in windows. Schemas live in a Finder that shows live vitals — row counts, freshness, storage tier, a time-travel sparkline — not a dead tree of names. Arrange an investigation across the canvas, save the whole desktop as a Scene, and reopen it tomorrow exactly where your thinking left off.
SQL blocks reference each other with {block} and re-run reactively when upstream results change — lineage included. Drop results on the chart shelf, compose a dashboard, publish it at a URL. The whole pipeline is SQL you can read.
Data Rabbit ships an assistant and speaks MCPboth ways — your agents (Claude, or your own) connect to the warehouse’s tools: search data by what it’s about, run governed read-only SQL, build dashboards, define metrics. Every model-backed answer carries a receipt naming the exact model version that produced it. Trust is a column, not a vibe.
Under the desktop sits RVBBIT— a Postgres 18 extension, not a hosted silo. Columnar acceleration with an adaptive router, table time travel, semantic operators you call in a WHERE clause, models trained and served from SQL. No lock-in: it’s your database, and it stays yours.
train_model() → predict_*(). AutoML as a SQL verb.The point of an integrated system isn’t a longer feature list — it’s that the pieces start cooperating on their own. From real deployments, lightly anonymized.
“Oh — someone on your team built something similar yesterday. Want to start from theirs?”
An analyst was sketching a dashboard when the assistant volunteered a colleague’s work from the day before. Nobody programmed that behavior.
memory × catalog × receipts
WHERE the witness statement sounds like a hoax
A WHERE clause that reads five thousand free-text reports and takes a position on each one — batched on GPUs, cached, and receipted, in one query.
semantic ops × acceleration × receipts
“The metric moved overnight. I diffed the table against yesterday before my coffee was done.”
Time travel isn’t a backup feature — it’s an argument-settling feature. Scrub back, diff, paste the offending rows into standup.
time travel × metrics × the scrubber
Data Rabbit is a great SQL desktop for the Postgres you already have. Put pg_rvbbit underneath and the same desktop gains a semantic, time-traveling, self-accelerating warehouse.
means(), about(), sentiment, extractionThe full stack — Postgres 18 with RVBBIT, plus Data Rabbit — or just the desktop pointed at a database you already run.
# The whole thing: Postgres 18 + RVBBIT + Data Rabbit curl -fsSL https://rvbbit.ai/install.sh | bash # Just the desktop, against your existing Postgres docker run -p 3000:3000 \ -e RVBBIT_LENS_SEED_DSN=postgresql://user:pass@host:5432/db \ ghcr.io/ryrobes/rvbbit-lens:latest
Both are plain Docker on your machine. Nothing phones home; connections and state live in a local volume. Docs, benchmarks, and the extension itself: rvbbit.ai.