A fast, open-source PostgreSQL GUI client

HeidiSQL's speed with TablePlus's polish — PostgreSQL-first, from the ground up. Native AOT startup in the ~100 ms range, streaming results, zero telemetry. Built with .NET 10 + Avalonia. MIT licensed.

Get it from Microsoft Store Download from Releases View on GitHub

or winget install pgNimbus --source msstore

pgNimbus main window: schema tree sidebar, SQL editor with completion, streaming results grid

Why pgNimbus

The gap in the Postgres client market: truly fast + open source + modern UI. pgAdmin and DBeaver are heavy; TablePlus is closed and paid; HeidiSQL is dated and MySQL-first.

Instantly fast

A NativeAOT binary that's on screen in the ~100 ms range — not an Electron shell. Startup, memory, and query-engine numbers are measured on every release.

PostgreSQL-first

Deep pg_catalog introspection, graphical EXPLAIN, FK-aware JOIN completion, LISTEN/NOTIFY, server activity — never the lowest-common-denominator dialect.

Keyboard-first

A Ctrl+K command palette that fuzzy-jumps to any table, query, or action; run, cancel, and navigate without touching the mouse. F1 shows every binding.

Private & open

Zero telemetry — no analytics, no crash reporting, no update pings. Passwords live in the OS credential store. MIT-licensed, so you can verify it, not trust it.

See it in action

Cold start to first frame, completion that predicts the next move, and the tools around the query.

pgNimbus launching from a cold NativeAOT process to a fully rendered main window in well under a second — Instant launch — cold NativeAOT process to rendered window

SQL completion: FROM with a partial table name, JOIN with an FK-ranked suggestion, ON auto-completing the full join condition — FK-aware completion — `JOIN` ranks related tables, `ON` completes the whole condition

Graphical EXPLAIN ANALYZE plan tree with per-node cost and timing — EXPLAIN ANALYZE as a plan tree, with per-node cost and timing

Command palette fuzzy-jumping to a table — Command palette — every table, saved query, and action behind Ctrl+K

Server activity window showing a live backend and its wait event — Live `pg_stat_activity` — cancel or terminate a runaway backend in one click

Connection dialog with saved profiles and paste-anything import — Paste-anything connections — URI, JDBC, libpq, even a full psql command line

Everything you need, nothing you don't

A minimalist shell over a serious engine — secondary actions live in the palette, not on a toolbar.

Streaming, cancellable results — the first screenful renders before the full result set arrives, in a virtualized grid.
Safe mode — stage grid edits, inserts, and deletes locally, review the exact SQL, commit as one transaction — or discard it all.
Schema-aware SQL completion — tables after FROM/JOIN, scoped columns, aliases, CTE output columns, user-defined functions.
Follow foreign keys from the grid — jump to the referenced row, or list every row referencing this one.
No-SQL table browsing — paged, sortable browsing pushed down to Postgres; the composed SQL doubles as your filter.
Transaction control — explicit Begin/Commit/Rollback on one held connection, with auto-rollback on failure.
Multi-tab editor — SQL formatting, find & replace, run history, saved queries, whole-script execution.
Grid CRUD — inline edits, add-row dialog, delete with confirmation, set-to-NULL — keyed on the primary key.
Cell inspector — double-click any cell to read a long text/jsonb value, JSON pretty-printed.
CSV/JSON import & export — type-inferred imports streamed via COPY; copy results as CSV, JSON, Markdown, or INSERTs.
Connection manager — per-connection accent colors, SSH tunnels, passwords in the OS credential store only.
LISTEN/NOTIFY monitor — subscribe to channels and watch notifications arrive live.

Fast is measured, not asserted

Every tagged release runs the benchmark suite: startup to first frame, memory at first frame, connect and round-trip latency, time to first row batch, and full-stream throughput of a 100,000-row query.

~100 ms

launch to first frame (NativeAOT)

100k rows

streamed while the first screenful is already visible

telemetry calls, ever

View the benchmark history →