Install Narwhal from PyPI

narwhal-inference installs the narwhal Python package and six commands for local development, engine launch, attestation, profiling, preflight and routing. Use Python 3.11 or newer on Linux.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install narwhal-inference
narwhal-serve --version
narwhal --help

narwhal-serve --version prints narwhal-inference and the distribution version from the Python environment behind the executable selected by the shell. Record that version with the fleet configuration, engine image, and profiles used for a deployment. Pin the approved version when installing on another router host.

narwhal --help lists the six installed commands and their purposes. Each command accepts --version before its operational arguments and exits with status 0. Source checkouts whose environment lacks distribution metadata print narwhal-inference unknown (distribution metadata unavailable); install the checkout with python -m pip install -e . to register its version.

Narwhal's wheel supplies the router commands. A production fleet also needs separately provisioned vLLM engines with compatible KV transfer, engine attestation, a fleet configuration, and measured profiles. Follow Deploy a fleet to qualify those inputs before serving requests.

The repository checkout carries deployment helpers and development checks. Run make setup in that checkout to install Narwhal and its development dependencies into .venv.