Installation

abkit ships as a single Python package. Installing it gives you three things:

Name	What it is
ab-analysis-kit	the pip package you install
abkit	the import package (import abkit, e.g. the pure abkit.stats core)
abk	the terminal command (abk init, abk run, abk explore, …)

Requirements

• Python 3.10 or newer (3.10, 3.11, and 3.12 are tested).
• A SQL warehouse to analyze against — ClickHouse (the first-class target), PostgreSQL, or MySQL 8.0.19+. The database driver is an optional extra (see Database drivers below) so abkit stays warehouse-agnostic and never forces a driver you don’t use.

You can install and use the pure statistical core (abkit.stats) and the config-lint path (abk run --steps validate) with no database at all — a driver is only needed once you point abkit at a real warehouse.

Install

Work inside a virtual environment so abkit and its dependencies stay isolated from your system Python:

python3 -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate

pip install ab-analysis-kit

That pulls in the runtime core automatically — numpy, scipy, statsmodels (the stats engine), pydantic + pyyaml (config), jinja2 (metric SQL templating), click (the CLI), plus orjson and requests. No database driver is installed by the base package.

Prefer a global CLI? If you only want the abk command available everywhere (not to import abkit inside another project), pipx works too: pipx install ab-analysis-kit. Add extras the same way, e.g. pipx install "ab-analysis-kit[all-db]".

Database drivers

Pick the extra that matches your warehouse. The extra names come straight from the package’s [project.optional-dependencies]:

Install command	Extra	Driver it adds
pip install "ab-analysis-kit[clickhouse]"	clickhouse	clickhouse-driver
pip install "ab-analysis-kit[postgres]"	postgres	psycopg2-binary
pip install "ab-analysis-kit[mysql]"	mysql	pymysql
pip install "ab-analysis-kit[all-db]"	all-db	all three: clickhouse-driver, psycopg2-binary, pymysql

The PostgreSQL extra uses psycopg2-binary, so you don’t need a C compiler or libpq headers to install it.

Keep the quotes around "ab-analysis-kit[...]" — some shells (zsh) treat the square brackets as globbing otherwise.

You only need the driver for the backend you actually connect to. If you’re not sure which to pick yet, [all-db] installs all three and lets you switch backends by editing profiles.yml later.

Other extras

Install command	Extra	When you want it
pip install "ab-analysis-kit[orchestration]"	orchestration	run abkit on a schedule with Prefect 3 (abkit never imports Prefect itself — the CLI is the unit of automation).
pip install "ab-analysis-kit[all]"	all	every runtime extra: all three DB drivers plus Prefect.
pip install "ab-analysis-kit[dev]"	dev	contributing to abkit — pytest, coverage/mock plugins, and pinned black/mypy/ruff.
pip install "ab-analysis-kit[integration]"	integration	run the Docker-backed integration tests (testcontainers for ClickHouse/PostgreSQL/MySQL).

Extras combine — separate them with commas, e.g. pip install "ab-analysis-kit[clickhouse,orchestration]".

Verify the install

Check that the abk command is on your PATH and reports a version:

abk --version

abk, version 0.0.1.dev0

List the available commands to confirm the full surface installed cleanly:

abk --help

You should see the shipped commands: init, init-claude, run, explore, validate, plan, unlock, and clean. Every command runs its body lazily, so abk --version and abk --help work instantly and require no database driver.

If abk isn’t found, your virtual environment probably isn’t active (re-run the source .venv/bin/activate step), or your PATH doesn’t include the environment’s bin/.

Install from source

To track the latest development build or to hack on abkit itself, install from a clone in editable mode with the dev extras:

git clone https://github.com/alexeiveselov92/ab-analysis-kit.git
cd ab-analysis-kit
pip install -e ".[dev]"          # add all-db for database work: ".[dev,all-db]"

The editable install exposes the same abk command; source edits take effect without reinstalling. See the contributor guide if you plan to change code.

Next steps

• Quickstart — scaffold a project with abk init and get your first result from abk run.
• Databases and Configuration — wire profiles.yml up to your warehouse (secrets stay out of YAML via {{ env_var('VAR') }} / ${VAR} interpolation).

abk init scaffolds a runnable example, so you can go straight from install to a real readout — no database required for the config lint (abk run --steps validate).