Command line & REPL

The dataglot binary is also a command-line SQL client. dataglot query and dataglot shell run the full engine in-process — federation, plan-time masking and row filters, the pg_catalog overlay — without starting a server or opening a port. Install the binary, point it at a config, and you're querying.

dataglot query -c dataglot.toml "SELECT * FROM pg.public.users LIMIT 5"

dataglot query — one statement, then exit

dataglot query [SQL] [-f <file>] [--format table|csv|json] [--user <name>] [-c <config>]

The SQL can come from three places (first match wins):

1. --file report.sql — read from a file
2. a positional argument — dataglot query "SELECT 1"
3. stdin — pass -, omit the argument, or just pipe:

echo "SELECT count(*) FROM pg.public.orders" | dataglot query -c dataglot.toml -

Results go to stdout; everything else (status, errors) goes to stderr, so output pipes cleanly. The process exits 0 on success and non-zero on any failure — a bad statement, an unreachable catalog, or no SQL provided — which makes it safe to use in scripts and CI with set -e.

Governance is not optional on this path: the embedded session applies the same plan-time masks and row filters the server enforces, so a one-shot extract can never leak what a psql session couldn't see.

Output formats

dataglot query -c dataglot.toml --format json "SELECT id, email FROM users" | jq .email

Flags

Config resolution follows the same precedence everywhere: CLI flag > env var > config file > default.

dataglot shell — the REPL

dataglot shell -c dataglot.toml

An interactive shell over the same embedded engine. Type a SQL statement on one line and press Enter; quit with \q, quit, exit, or Ctrl-D.

dataglot shell — type SQL and press Enter; \q or Ctrl-D to quit.
dataglot> SELECT u.email, o.amount FROM users u JOIN pg_orders.public.orders o ON u.id = o.user_id;

• Prompts print to stderr, results to stdout — so even the REPL can be driven by a script and its output captured cleanly.
• A statement error prints and the shell keeps going; only a fatal stdin error ends it.
• --format and --user work exactly as in query.

dataglot init — scaffold a config

dataglot init                # writes ./dataglot.toml
dataglot init custom.toml    # custom path
dataglot init --force        # overwrite an existing file

Writes a commented starter config (one Postgres catalog and a mask example) and prints the next steps. It refuses to overwrite an existing file unless --force. dataglot --print-example-config streams the same content to stdout instead, for piping or inspection.

Shell completions

dataglot completions bash > /etc/bash_completion.d/dataglot

dataglot completions zsh > "${fpath[1]}/_dataglot"

dataglot completions fish > ~/.config/fish/completions/dataglot.fish

Health probe

dataglot --healthcheck performs a fast TCP connect to the server port (default 5432, --port to override) and exits 0/1 — the same contract as nc -z, but available inside shell-less container images. It's what the container image's HEALTHCHECK uses.

Where next

• Data pipelines from the command line — composing dataglot query with jq, csvkit, psql, cron, and CI.
• Configuration file — the full dataglot.toml reference the CLI loads.
• Runtime configuration — the SQL DDL surface (CREATE CATALOG, CREATE MASK, …). Note: runtime DDL is handled on the server's wire-protocol path, so it needs a running server and a Postgres client — the in-process query/shell sessions read their sources from the config file.