Install¶
Kernos is four components. Install the ones you run; a laptop needs all four to try it, a production deployment usually runs the kernel and gateway as services and the workers wherever the model traffic should originate.
From the registries¶
On crates.io, alongside the two
libraries it is built from,
kernos-core and
kernos-policy, which you can
depend on separately. Or download a binary for Linux or macOS from the
releases page.
Documented at
pkg.go.dev, including
the connect package for writing your own connector. Binaries for Linux
and macOS are attached to every
release.
pip install "kernos-sdk[anthropic]" # the extra pulls the Anthropic SDK
kernos-worker --help
kernos-eval --help
On PyPI. Python 3.10 or newer.
On npm, published with
provenance. Node 18 or newer, or any browser with fetch.
docker pull ghcr.io/rhs2/kernos-kernel:latest
docker pull ghcr.io/rhs2/kernos-gateway:latest
docker pull ghcr.io/rhs2/kernos-worker:latest
Listed under the repository's packages.
deploy/docker-compose.yml in the repository starts all three with the
reference connectors.
From source¶
git clone https://github.com/rhs2/kernos && cd kernos
make build # target/release/kernos, gateway/bin/kernos-gateway, sdk/python/.venv, sdk/typescript/dist
make test # unit and integration tests in all four languages
make accept # the end-to-end acceptance suite (no API key, no network)
Toolchains: Rust stable, Go 1.22 or newer, Python 3.10 or newer, Node 18 or
newer. make prints every target with make help.
Model access¶
Nothing needs an API key to run: the mock provider executes any bundle end to
end from the mock outputs the bundle declares, which is how the acceptance
suite and the evaluation harness run in CI. To use real models, set
ANTHROPIC_API_KEY and start workers with --provider anthropic.