Installation#

sktime-cli runs on Python 3.10 to 3.14 and installs sktime as a dependency.

Install the CLI#

To install sktime-cli as a standalone tool with its own isolated environment, use uv:

uv tool install sktime-cli

To install it into the active environment instead, use pip:

pip install sktime-cli

Both commands provide one executable, sktime-cli. There is no short sktime alias, which keeps the command from colliding with the library.

Check the installation#

doctor reports whether sktime imports, whether the cache directory is writable, and which optional dependencies are available:

sktime-cli doctor

The output is similar to the following:

Output of sktime-cli doctor, listing sktime version, cache state, and optional dependencies

doctor exits with 0 even when optional dependencies are missing. It exits with 1 only when sktime itself fails to import.

Optional dependencies#

sktime ships most of its estimators with optional dependencies, and the CLI adds one extra of its own. Nothing is installed by default.

parquet

Adds pyarrow so that data convert, data inspect, and datasets load can read and write parquet files.

uv tool install "sktime-cli[parquet]"
Estimator dependencies

Estimators such as AutoARIMA and Prophet need packages that sktime doesn’t install. When a command needs one, it exits with code 3 and a hint that names the install command. To see what an estimator needs before you run it, use registry describe NAME, or filter searches down to what you can already run with registry search --installable-only.

Install from source#

To work on the CLI, clone the repository and sync the development environment:

git clone https://github.com/siddharth7113/sktime-cli
cd sktime-cli
uv sync

This resolves sktime from PyPI, like any other dependency.

For the full development workflow, see Contributing.