Command line

sqldatapack does what the library does, without a C# project. Export a slice, edit the file, import it back.

This is the right tool when the job is one-shot: cutting a dev slice, handing someone a bug repro, running an export from a scheduled task. If you want the behaviour inside an application, use the library instead.

Install

The Windows builds and the direct downloads carry their own .NET runtime, so nothing needs to be installed first.

winget install zachtbeer.SqlDataPack

On a shared server, install for every user rather than just yours:

winget install zachtbeer.SqlDataPack --scope machine

If you already have the .NET SDK:

dotnet tool install -g SqlDataPack.Cli

Or download a single executable from the releases page for win-x64, win-arm64, linux-x64, linux-arm64, osx-x64 or osx-arm64. Each release also carries a SHA256SUMS file and a build provenance attestation.

curl -sSL https://github.com/zachtbeer/sqldatapack/releases/latest/download/sqldatapack-linux-x64.tar.gz | tar xz
./sqldatapack --version

Export

sqldatapack export \
  --connection "Server=.;Database=Northwind;Integrated Security=true" \
  --out dev-slice.sqlite \
  --tables dbo.Customers,dbo.Orders,dbo.Invoices \
  --exclude-column dbo.Customers.NationalId \
  --global-where "CustomerId:CustomerId = 42"

Exported 1,847 rows from 3 tables into dev-slice.sqlite

Flag	What it does
--connection, -c	SQL Server connection string. Falls back to SQLDATAPACK_CONNECTION.
--out, -o	Path of the SQLite package to write. Required.
--tables	Export only these tables. Repeat the flag or separate with commas.
--exclude-tables	Export everything except these. Cannot be combined with --tables.
--exclude-column	Leave a column out entirely, as schema.Table.Column. Repeatable.
--global-where	"Column:predicate", applied to every table that has the column. Repeatable.
--table-where	"schema.Table:predicate", applied to one table. Repeatable.
--schema	none (default) or dacpac, to capture the source schema alongside the data.
--overwrite	Replace the output file if it exists.
--batch-size	Rows per batch. Defaults to 1,000.
--timeout	SQL command timeout in seconds.
--options	JSON file for everything else. See below.
--quiet, -q	Suppress per-table progress. Warnings still print.

An excluded column is never read from SQL Server, so the data does not cross the wire at all. A global WHERE fails open: a table without that column exports unfiltered and records a warning. See Options.

The predicate is split from the key on the first colon, so a predicate can contain one:

sqldatapack export -c "..." -o slice.sqlite --global-where "CreatedAt:CreatedAt > '2024-01-01 08:30:00'"

Import

sqldatapack import dev-slice.sqlite \
  --connection "Server=.;Database=NorthwindDev;Integrated Security=true"

Flag	What it does
--connection, -c	SQL Server connection string. Falls back to SQLDATAPACK_CONNECTION.
--deploy-schema	none (default) or dacpac. Needs a package exported with --schema dacpac.
--row-count-drift	warn (default) imports what the file holds; fail rejects a package whose counts moved.
--batch-size	Rows per batch. Defaults to 1,000.
--timeout	Bulk copy timeout in seconds.
--options	JSON file for everything else.
--quiet, -q	Suppress per-table progress.

The target tables must already exist and be empty, unless the package carries a dacpac and you pass --deploy-schema dacpac. Editing the file changes its row counts, which is normal; --row-count-drift fail is what an unattended scrubbing pipeline wants. See Importing.

Keeping the connection string out of the command

Anything on the command line lands in shell history and in the process list. Use the environment variable instead:

export SQLDATAPACK_CONNECTION="Server=.;Database=Northwind;Integrated Security=true"
sqldatapack export --out dev-slice.sqlite --tables dbo.Customers

The options file

The flags cover the common path. Everything else lives in a JSON file whose property names are the library's own option names:

{
  "tableSelection": "Only",
  "tables": ["dbo.Customers", "dbo.Orders"],
  "excludeColumns": ["dbo.Customers.NationalId"],
  "globalWhereClauses": [
    { "columnName": "CustomerId", "whereClause": "CustomerId = 42" }
  ],
  "schemaCaptureMode": "Dacpac",
  "dacpacCaptureOptions": { "schemaScope": "SelectedExportTables" },
  "batchSize": 5000
}

sqldatapack export -c "..." -o dev-slice.sqlite --options support-slice.json

Three things worth knowing:

• Explicit flags win. --batch-size 250 overrides the file. A flag you did not type changes nothing, so a file can describe a slice while one run tweaks a single value.
• An unknown property is an error, not something silently ignored. A typo that quietly did nothing would produce a slice that is wrong in a way nobody notices.
• A connection string in the file is refused. The file is meant to be committed and reviewed; credentials are not. Keep them on the command line or in SQLDATAPACK_CONNECTION.

Comments and trailing commas are allowed.

Exit codes

Code	Meaning
0	Done.
1	SqlDataPack refused the operation. The message prints without a stack trace.
2	The command line or the options file was wrong.
3	Something unexpected. Re-run with --verbose for a stack trace.
130	Cancelled with Ctrl+C.

Progress goes to stderr and the result summary to stdout, so sqldatapack export ... > result.txt keeps the two apart.

Locked-down machines

The downloadable and winget builds are one self-contained file that unpacks itself into %TEMP%\.net\ ($TMPDIR on Linux and macOS) the first time each version runs. That costs a couple of seconds once per machine, and it is why the tool needs no .NET installed.

Where policy blocks execution out of the temp directory, point it somewhere allowed:

$env:DOTNET_BUNDLE_EXTRACT_BASE_DIR = "C:\ProgramData\SqlDataPack\bundle"
sqldatapack --version

The binaries are not code signed. winget verifies the SHA256 from the manifest, and every release carries a build provenance attestation you can check with the GitHub CLI:

gh attestation verify sqldatapack-win-x64.exe --repo zachtbeer/sqldatapack

What the CLI does not do

inspect and preflight are not here yet. To read a package's manifest without touching a database, or to find out what an import would reject before it writes anything, use SqlDataPackReader.ReadManifestAsync and PreflightAsync from the library.