Installation

sekai needs Python 3.10 or newer. The core package depends only on NumPy (≥ 1.24) and cloudpickle (used to ship environment factories to ProcVecEnv workers).

pip install sekai

Check the install with the command-line tool:

sekai --version
sekai list

Optional extras

Extra	Installs	Needed for
sekai[gymnasium]	gymnasium>=1.0	sekai.compat: to_gymnasium() / from_gymnasium(), Stable-Baselines3, CleanRL, RLlib
sekai[render]	pillow, imageio, imageio-ffmpeg	saving frames: save_gif() (Pillow), save_video() (imageio + ffmpeg)
sekai[pettingzoo]	pettingzoo>=1.24, gymnasium	to_pettingzoo() / from_pettingzoo() for multi-agent libraries
sekai[pandas]	pandas>=1.5	MarketData.from_pandas() / from_long()
sekai[all]	gymnasium, pettingzoo, pandas and render extras	everything at runtime
sekai[test]	pytest, pytest-cov, gymnasium, pettingzoo, pandas, pillow	running the test suite
sekai[docs]	mkdocs-material, mkdocstrings[python]	building this site
sekai[dev]	test + docs + mypy, ruff, pre-commit, build, twine	contributing

pip install "sekai[gymnasium,render]"

Rendering needs no extras

Drawing frames uses sekai's own NumPy Canvas, so CartPoleRenderer().render(env) works with a bare install. The render extra is only for writing GIF/MP4 files.

From source

git clone https://github.com/QianCapital/sekai
cd sekai
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e ".[dev,gymnasium,render]"
pytest

See Contributing for the full development workflow.

Type checking

sekai ships a py.typed marker and type-checks clean under mypy --strict. Environments, spaces and results are generic, so annotations like sekai.Env[np.ndarray, np.intp] and sekai.StepResult[np.ndarray] flow through to your code.