A fast PostgreSQL client that talks to your database and nothing else.
On screen in about 0.2 s. Results stream while the query runs. No telemetry, no account, no cloud.

Website · Docs · Install · Features · Roadmap

pgNimbus is written by a developer who spends most working days in Postgres, for people who do the same. Three things don't bend:

• It's fast where you wait. A NativeAOT binary, not Electron: about 0.2 s to a window, rows on screen while the query is still running, and Esc stops a query mid-flight. Measured on every release.
• Your data stays yours. No telemetry, no update checks, no crash uploads, no account, no AI features. The only connections it opens go to your Postgres servers and SSH tunnels.
• It's built for the keyboard. Ctrl+K reaches any table or command, a connection string in any format fills the form, and autocomplete reads SQL the way the server does.

Where it sits among the alternatives: pgAdmin and DBeaver are powerful but heavy. TablePlus is fast and polished, but closed source and paid. Beekeeper Studio is open source but runs on Electron. HeidiSQL is native and fast, but dated and MySQL-first.

And next to an AI agent? pgNimbus itself is built with Claude Code, so this isn't a knock on agents. Let the agent write the hard query. Looking at the rows, fixing one value in production, or finding who holds a lock right now is quicker in a client, and the rows never end up in a model's context.

Cold start (NativeAOT)	Completion that knows your foreign keys

EXPLAIN ANALYZE as a tree	Safe mode: review, then commit once

The installation guide has the step-by-step for each platform, the macOS Gatekeeper dialogs, and where pgNimbus keeps its files. Every release asset carries signed build provenance, so you can check where a download came from:

gh attestation verify pgNimbus-<version>-win-x64.msi --repo Shman4ik/pgNimbus

1. Launch pgNimbus. The connection dialog opens first.
2. Paste a connection string into the box at the top. A postgres:// URI, JDBC URL, Host=…;Port=…, libpq keywords and a whole psql command line all work.
3. Connect. The form saves itself as you type; there is no Save button. Next time, launch and press Enter to reconnect.
4. Run with Ctrl+Enter, jump anywhere with Ctrl+K, and press F1 for every shortcut.

On macOS, Cmd replaces Ctrl, except autocomplete, which stays on Ctrl+Space. More in Connecting to a database.

SQL editor (guide)

• Autocomplete that resolves names the way the server does: along search_path, per query block (subqueries, CTEs, LATERAL), with one JOIN … ON condition offered per foreign key, argument hints for functions and only types after ::.
• Formatting that only ever changes whitespace, ;-separated scripts with a result section per statement, find and replace, .sql files, searchable query history and saved queries.

Results and editing (guide)

• Streaming, virtualized grid. Browse a table without SQL; sorting, paging and filter chips run on the server.
• Safe mode: edits, inserts and deletes are staged, you review the exact SQL, and everything commits as one transaction. At commit each staged row is re-read under a lock, and if another session changed it the batch rolls back and shows before, current and proposed values.
• Type-aware editors (enum dropdowns, booleans, dates, arrays, composites, json/jsonb), a row-details form, a cell inspector, and follow-the-foreign-key from any key cell.
• CSV/JSON import through COPY; copy results as CSV, TSV, JSON, Markdown or INSERT statements.

PostgreSQL tooling (plans, monitoring)

• Schema tree read from pg_catalog: materialized views, partitioned tables, relation sizes, DDL for any table or view.
• EXPLAIN and EXPLAIN ANALYZE as a tree with a self-time heat map and plain warnings (disk spills, bad row estimates). Paste a plan from anywhere and read it with no connection.
• Server activity with cancel and terminate, plus a who-blocks-whom lock tree from pg_blocking_pids.
• Database overview: largest relations, unused indexes, seq vs index scans, cache hit ratios.
• Slow queries from pg_stat_statements: ranked by total or mean time, measured since the last reset or over just the workload you ran, one double-click from the editor.
• Roles and permissions that answer "can this role do that, and why" from the server's own has_*_privilege(), including grants inherited through roles and PUBLIC. Changes come out as a script, never applied behind your back.
• LISTEN/NOTIFY monitor with JSON payloads as a tree and a button to publish a test event.

Connections (guide)

• Saved profiles with accent colours, so production never looks like staging.
• SSH tunnels with agent, key file or password auth.
• Several databases side by side, each window with its own pool and tunnel. Auto-reconnect after sleep, without ever silently re-opening a transaction.
• Tabs, including unsaved ones, come back after a restart.

Every tagged release runs a benchmark job and publishes the history at https://shman4ik.github.io/pgNimbus/dev/bench/. At v0.13.1, on a GitHub Ubuntu runner:

The numbers are machine-relative; the point of the chart is that a regression shows up as a step in the release that caused it. scripts/benchmarks/run-benchmarks.sh runs the same suite locally.

pgNimbus sends zero telemetry: no usage analytics, no automatic crash uploads, no update pings. There's no account and no cloud sync, and nothing you query, browse or type is sent anywhere except the servers you configure. Saved passwords go to the OS store (DPAPI on Windows, Keychain on macOS, Secret Service on Linux), never into the profile file.

Two things to know. Query history, workspace SQL and the local crash log can contain sensitive data and are stored on disk unencrypted. And if the OS store is unavailable, a warning says so and the password is kept only for the current session. Details, including migration from older unencrypted files, are in where your password goes.

Building from source is the one place anything is reported, and it isn't the app: Avalonia's build tooling sends anonymous build statistics to Avalonia while the project compiles. Nothing of it ships in the binaries you download.

Next up (a direction, not a commitment):

• Production connection policies: environment labels, on top of the read-only sessions that shipped.
• Query Lab: save EXPLAIN runs and compare them before and after a change.
• Signed and notarized macOS builds.

The full backlog, with the competitive research behind each item, is in ROADMAP.md. Pick one scoped item if you'd like to contribute; CONTRIBUTING.md has the details.

Requires the .NET 10 SDK.

dotnet build
dotnet run --project PgNimbus.App
dotnet test --project PgNimbus.Core.Tests

PgNimbus.Core is the engine: a plain class library over Npgsql with no UI dependencies, streaming results as IAsyncEnumerable<RowBatch> with real mid-flight cancellation. PgNimbus.App is the Avalonia front end. To skip the connection dialog while developing, set PGNIMBUS_CONN to any connection string the paste box accepts.

A NativeAOT build:

dotnet publish PgNimbus.App -c Release -r win-x64 -p:PublishAot=true    # Windows
dotnet publish PgNimbus.App -c Release -r linux-x64 -p:PublishAot=true  # Linux (needs clang + zlib1g-dev)

The docs site is MkDocs Material: pip install -r docs/requirements.txt && mkdocs serve.

MIT, see LICENSE.