conda-ship

Build ready-to-run conda runtimes from solved conda environments.

conda-ship builds executables that install a locked conda environment on first use, then run a command from that environment. The installation directory is called the managed prefix. The command, usually conda or python, is the delegate.

The package provides the cs builder CLI, the cs-template runtime template, and a Python adapter for conda ship. A composite GitHub Action builds runtimes in CI. The project is currently alpha and pre-1.0.

Downstream distributions choose their package sets, runtime names, release channels, installers, and user documentation. For example, conda-express uses conda-ship to publish the cx and cxz runtimes.

Quickstart

This example uses conda-workspaces to create a solved source environment, then builds an online runtime named demo:

conda install --name base -c conda-forge conda-pypi
conda create -n cs-demo -c conda-forge python pip conda-workspaces
conda activate cs-demo
conda pypi install conda-ship

mkdir demo-runtime
cd demo-runtime
conda workspace init --format conda --name demo-runtime
conda workspace add --feature ship --no-lockfile-update \
  "python>=3.12" \
  "conda>=25.1" \
  conda-rattler-solver \
  "conda-spawn>=0.1.0"
cat >> conda.toml <<'TOML'

[tool.conda-ship]
runtime-name = "demo"
runtime-version = "0.1.0"
delegate-executable = "conda"
artifact-layout = "online"
source-environment = "ship"
exclude-packages = ["conda-libmamba-solver"]
TOML

conda workspace lock
cs inspect
cs build --dry-run
cs build
./dist/demo info

For a guided walkthrough with automatic first-run bootstrap and embedded runtime examples, see the first runtime tutorial.

What It Builds

conda-ship stages a runtime binary plus release metadata:

• .runtime.lock: the lockfile stamped into the runtime
• .packages.txt: tab-separated package records for quick inspection
• .cdx.json: CycloneDX 1.7 SBOM for the resolved conda package graph
• .info.json: artifact metadata for release tooling
• .sha256: checksums for staged files
• optional .bundle.tar.zst: compressed package archives for offline builds

On first invocation, the runtime automatically bootstraps its managed prefix. It then passes every argument to the configured delegate executable, usually conda. The delegate and its plugins handle --help, --version, and every subcommand. conda-ship does not reserve those arguments.

The finished executable contains the native bootstrap and optional update code from cs-template. Neither the executable nor its managed prefix needs the conda-ship Python package.

During bootstrap, generated runtimes also write constructor-compatible conda prefix metadata. The managed prefix gets conda-meta/history and conda-meta/initial-state.explicit.txt in addition to conda-ship's ownership metadata. Conda can recognize the prefix as an environment, and runtimes that include conda-self can use that initial-state snapshot for conda self reset --snapshot installer-updated or conda self reset --snapshot installer-exact.

Artifact Layouts

Project Input

conda-ship builds from an already solved source environment. It does not solve loose matchspecs in the GitHub Action and it does not define a package set of its own.

Supported manifest and lockfile pairs:

• conda.toml plus conda.lock
• pyproject.toml with nonempty [tool.conda.workspace] plus conda.lock
• pixi.toml plus pixi.lock
• pyproject.toml with nonempty [tool.pixi.workspace] plus pixi.lock

The selected source environment defines the packages and channels. [tool.conda-ship] configures the runtime build:

[tool.conda-ship]
runtime-name = "demo"
runtime-version = "0.1.0"
delegate-executable = "conda"
artifact-layout = "online"
source-environment = "ship"
exclude-packages = ["conda-libmamba-solver"]

The selected source environment defines the complete package set. It must include the configured delegate executable and any optional commands the distribution exposes. A conda distribution can include conda, conda-rattler-solver, and conda-spawn, while a runtime with another delegate can omit them. Include conda-self when a conda runtime should expose conda self reset for restoring the bootstrapped base prefix to the initial package set shipped by the runtime.

conda-ship does not choose conda configuration. Set condarc-file to copy a YAML condarc file to <prefix>/.condarc, and set freeze-base = true when it should write a CEP 22 frozen marker. Both are disabled by default.

Local Workflow

Packaged builds find cs-template next to the installed cs executable. Use --template only for an explicit template path, custom packaging, or cross-builds.

cs inspect
cs build --dry-run
cs build
cs build --artifact-layout embedded
cs run --install-path /tmp/demo-smoke -- info

cs inspect is the preflight command. It derives the runtime lock, validates the selected source environment, applies package exclusions, and prints the package set without writing artifacts.

Use cs build --dry-run to preview the runtime metadata and staged release asset paths before writing files.

After a real build, verify the staged artifacts and inspect the release metadata before handing them to downstream packaging or signing.

The staged runtime is a stamped copy of the generic runtime template. It bootstraps the managed prefix if needed, then runs the configured delegate.

CONDA_SHIP_PREFIX overrides the managed prefix for every runtime. A runtime-specific variable such as DEMO_PREFIX remains available for other runtime names. A runtime named conda does not use CONDA_PREFIX for this purpose because that variable may describe an activated environment.

GitHub Actions

The repository root is also a composite GitHub Action for downstream release jobs:

- uses: conda-incubator/conda-ship@FULL_RELEASE_COMMIT_SHA # X.Y.Z
  id: cs
  with:
    conda-ship-version: "X.Y.Z"
    artifact-layout: embedded

The action expects a committed manifest and matching lockfile. It downloads the configured cs, cs-template, and SHA256SUMS release assets for the runner, verifies their GitHub Artifact Attestations, checks the release checksums, runs cs build --dry-run, and then stages the runtime into a dist-path output.

Pin the action source to a full release commit SHA for release builds and pass the matching conda-ship release through conda-ship-version. When the action is invoked by an exact release tag, conda-ship-version can be omitted for backwards compatibility.

Packaging

conda-ship is not an OS installer generator. It does not target .sh, .pkg, or .msi output directly. It produces runtimes that can be distributed as GitHub Release assets or wrapped by Homebrew, constructor, Docker, enterprise packaging systems, and other release tooling.

The PyPI package installs the cs builder, the cs-template runtime template, and the Python adapter together. The adapter makes conda ship a shortcut for the same builder when installed in a conda environment. It does not make conda-ship part of conda itself. A conda package for the builder should install the same pieces together in its builder environment.

What Belongs Downstream

Downstream distributions decide:

• runtime names and delegates
• package sets and channels
• package exclusions
• install schemes and install names
• documentation URLs
• release channels and installers
• signing, SBOM publication and retention policy, and in-toto provenance for final artifacts

conda-ship verifies the inputs it consumes and the package archives it stages or installs. Downstream release systems should first verify and, when required, attest the unchanged cs build output. They should then sign a copy of the runtime, verify its native signature, and publish a separate checksum or attestation for those finalized bytes. The original .sha256 and .info.json continue to describe the pre-sign build output.

Documentation

Full documentation is available at conda-incubator.github.io/conda-ship.

Useful starting points:

• Build your first runtime
• Build in GitHub Actions
• Configuration reference
• Project scope

Development

pixi install
pixi run test
pixi run lint
pixi run -e test pytest
pixi run docs

pixi run lint runs the repository's prek hooks, including Rust formatting and clippy checks. See the contributor coverage commands for Rust and Python reports.

The terminal demos are generated from demos/*.tape with VHS:

pixi run demos
pixi run demos inspect

The tapes build local debug cs and cs-template binaries in hidden setup so the visible commands match the packaged workflow.

Run cargo generate-lockfile after changing Cargo metadata and pixi lock after changing pixi metadata.

Maintenance

conda-ship is a federated project in conda-incubator. The conda-ship project team maintains this repository. For help, open an issue.

License

BSD-3-Clause