Generate data

Generate sample datasets from the terminal — or fetch real-world rows from APIs and public sources. Useful for tests, demos, prototypes, and quick data pulls.

Natural language

Synthetic (default)

arka generate 100 users as csv
arka generate sample sales data json --rows 50
arka generate fake emails tsv
arka create sample json data
arka data_gen --schema "name,email,age" --format json --count 20

Real-world sources

arka generate real world bank gdp india 2010-2024 as csv
arka generate 20 pubmed papers on "mRNA vaccines" as csv
arka generate data from https://pokeapi.co/api/v2/pokemon?limit=10 as jsonl

Phrases with real, actual, live, world bank, pubmed papers, or a URL route to real fetchers. Fake / sample / mock keep synthetic generation.

CLI commands

Command	Description
`generate_data users --count 100 --format csv`	User records preset (synthetic)
`generate_data --source worldbank --indicator gdp --country IN -f json`	World Bank GDP for India
`generate_data --source pubmed --query "mRNA vaccines" -n 20 -f csv`	PubMed paper metadata
`generate_data --source url --url "https://pokeapi.co/api/v2/pokemon?limit=10" -f jsonl`	JSON or CSV from URL
`generate_data --real --indicator population --country US`	Shorthand: infer World Bank
`generate_data --fields "id,name,price,category" -n 20`	Custom synthetic columns

Real-world sources

`--source`	Description	Key flags
`worldbank`	GDP, population, life expectancy, CO₂, unemployment, internet	`--indicator`, `--country`, `--year-from`, `--year-to`
`pubmed`	Biomedical paper metadata (title, journal, year, URL)	`--query`, `--count`
`url`	Fetch JSON array/object or CSV from any URL	`--url`, `--count`
`web`	DuckDuckGo search results as rows	`--query`, `--count`

Real sources cap at 500 rows. When --source is set, Arka never falls back to synthetic or LLM data — fetch errors are reported clearly.

World Bank indicators

gdp, gdp_per_capita, population, life_expectancy, co2, unemployment, internet Country codes use ISO-3166 alpha-2 (e.g. IN, US). Natural language country names (India, United States) are parsed automatically.

Supported formats

csv, json, jsonl, tsv, yaml, xml, sql, markdown, xlsx XLSX output requires openpyxl (pip install openpyxl). For richer fake values, install the optional Faker extra:

pip install faker

Presets

Built-in presets infer columns automatically (synthetic mode only):

Preset	Columns
`users`	id, name, email, age, phone
`products`	id, name, price, category
`sales`	date, product, quantity, revenue, region
`emails`	email
`customers`	id, name, email, address, phone

Field names are matched to generators (names, emails, phones, addresses, dates, numbers, UUIDs, booleans). Use --llm for complex custom schemas when an LLM API key is configured.

Output

Print to stdout by default, or save with -o output.csv or by passing a filename as the first argument.

Routing & CLI

Agent & memory

Developer tools

Models & inference

Media & documents

Integrations

Finance & data

Workflows & productivity

More

Generate data

Natural language

Synthetic (default)

Real-world sources

CLI commands

Real-world sources

World Bank indicators

Supported formats

Presets

Output

​Natural language

​Synthetic (default)

​Real-world sources

​CLI commands

​Real-world sources

​World Bank indicators

​Supported formats

​Presets

​Output

Natural language

Synthetic (default)

Real-world sources

CLI commands

Real-world sources

World Bank indicators

Supported formats

Presets

Output