CLI

The datagen CLI is the primary surface. Every workflow a human or a coding agent can drive — creating a dataset, reviewing a preview, registering a Resource, managing auth — is a subcommand. The web UI exists for humans who prefer to click through a preview; everything else lives here.

Command tree

datagen
├── datasets   # create, list, get, preview, approve, feedback, watch, download
├── resources  # list, get, create, delete, templates, materialize
├── auth       # login, logout, status, org
├── config     # show
└── health     # check API connectivity

Five groups. Four of them are noun-first (datagen <noun> <verb>). The fifth — health — is a top-level sanity check.

Progressive --help

Every subcommand is self-documenting. --help at any depth prints the flags, their types, their defaults, and a one-line description of what the command does. Point a coding agent at the CLI and it can resolve the next command from the current one without ever reading these docs:

datagen --help
datagen datasets --help
datagen datasets create --help

This is the recommended way for Claude Code, Cursor, or Windsurf to drive the CLI. See Claude Code workflow for the full pattern.

Output formats

Every subcommand accepts --format with two values:

Data goes to stdout; diagnostics and prompts go to stderr. You can safely pipe --format json output without interleaved noise.

Global options

Connection settings (API URL, token, timeout) come from environment variables and a credentials file that datagen auth login manages — see Config.

The health command

A top-level ping against the API. Useful as a first check after install or auth.

$ datagen health
Connected to https://api.datagen.example (12 models available)

Pass --format json for a structured payload.

Where to go next