DATA EXPLORER
The raw-SQL room. Pick any connection from your profile, browse its schemas and tables live, and run SQL on the engine itself— in that engine's own dialect, computed where the data lives. Where Chat with Martin answers questions through federation, the Explorer is for the moments you already know the query and just want a grid.
STARTING
dvt serve # the suite, explorer included dvt explorer # just the explorer (46107) dvt kill -s explorer # stop it
WHAT IT DOES
The connection picker lists every output in your profile — databases and warehouses alike. Selecting one loads its schema tree live (the same engine-correct metadata queries that power generate-sources); clicking a table drops a SELECT * … LIMIT 100into the editor. Queries run on the connection's own engine in its own dialect — no federation, no DuckDB in the middle — and the grid shows exactly what the engine returned, capped at 2,000 rows and honest about it.
THE GRID EDITOR
Browse to any table — a click in the tree, or a deep link like #pg_prod/public/orders (append /edit to land straight in edit mode) — and toggle ✎ EDIT. Change cells, add rows with + ROW, mark rows for deletion with ✕. Every pending change previews as the exact SQL that will run — one builder produces both the preview and the execution, so the preview cannot lie.
Rows are identified by the discovered primary key (marked 🔑) or, on engines without PK metadata, by every original column value — NULL-safe. A keyless UPDATE or DELETE is refused server-side. Nothing runs before an explicit APPLY, and per-statement affected counts surface keys that matched 0 rows — or 2 or more.
THE RAW-SQL BOX: READ-ONLY, ENFORCED
The SQL box is SELECT-only — and that is now enforced, not advisory: every statement is parsed, one SELECT per run, writes and DDL refused on sight, results always bounded. Mutations have two honest doors: the grid editor above, with its preview-exact SQL behind APPLY, and dvt exec in the terminal.