Home / Manual

The Pipewright manual

Everything you need to take a model from a notebook to a governed, monitored release on Google Cloud. Written for ML engineers, data scientists, SQL analysts and platform engineers, for Pipewright 1.3.0.

Quickstart →An environment, the CLI and your first released use case, in an afternoon. Concepts →Environments, manifests, use cases, releases, the launcher, gates and decisions: the why behind each.

Tutorials

Each tutorial builds on the one before, on a real environment. Each ends by telling you what it left running and what that costs.

1. Your first use caseThe churn example: validate, release, train, judge and approve. 2. Models in SQLSegments and a forecast in BigQuery ML, with no Python. 3. Serving and canariesAn endpoint, a canary, and features read from the online store. 4. Monitoring and retrainingDrift, alerts and a retrain case that a person decides. 5. Promotion and CI/CDFrom test to production through GitHub Actions, without keys. 6. Running on GKEKubeflow Pipelines, KubeRay and KServe in a private cluster. 7. Build with an agentClaude Code onboards a use case from the feature catalogue; you review, merge and approve.

AI build assistance

Coding agents draft and check use cases through Pipewright’s tools, as an identity that reads metadata only; a person approves every change. New in 1.3.0.

The guideThe agent’s identity, context packs, the tools, judgement checks, the review bundle and CI. On EnterpriseA hosted server per team, aggregates under rules, the audit, EV-16 and the platform kit. Things to watch out forAntivirus and TLS proxies, folder trust, ID tokens, cold starts, stale packs: symptom, cause, fix.

Reference

The use case filesEvery file in a use case folder and the settings that matter most. CommandsEvery pipewright command, by stage of the lifecycle. Google Cloud servicesWhat each service is, why it matters, and how Pipewright uses it. Design and decisions ↗CONTEXT.md in the repository: every interface and the reasoning behind it.
Help in your editor

Every YAML file pipewright new-usecase writes points at Pipewright’s JSON Schemas, so editors with YAML support suggest keys, explain them on hover and underline mistakes. pipewright explain <key> prints any key’s meaning, type, default and whether it may differ per environment.