Culvert

A cloud-agnostic, polyglot data-pipeline framework. One contract set — Source, Sink, Transform, Pipeline, RuntimeContext, BlobStore, Warehouse, JobControlRepository, and friends — realised in Java and Python, with cloud specifics behind adapters. Define the pipeline once against contracts; point it at an emulator on your laptop, a dev project, or production by wiring, not rewriting.

• Contract-driven — the language-neutral spec is the portability boundary. Business logic depends only on contracts.
• Two clouds today — GCP is the first full implementation; AWS is a real (Java) adapter family with the same pipeline proven on both. Azure is on the roadmap.
• No application framework in the core — plain Python Protocols + entry-point auto-discovery; composes into Beam pipelines, Airflow DAGs, or a script on your laptop.
• Local-first — the whole stack runs against emulators with zero cloud account; cloud is where you prove and ship, not where you develop.

Install

pip install culvert                 # core contracts only (no cloud SDKs)
pip install culvert[gcp]            # + BigQuery, GCS, Pub/Sub, Secret Manager, observability
pip install culvert[orchestration]  # + Airflow-side DAG factory, operators, sensors
pip install culvert[transform]      # + dbt integration
pip install culvert[all]

Python ≥ 3.10. For 0.1.0 the import packages keep their library names — the contracts live in data_pipeline_core:

from data_pipeline_core import autoconfig

config = autoconfig.discover()          # entry-point adapter discovery
blob_store = config.blob_store()        # GcsBlobStore if culvert[gcp] installed

The Java twin ships as com.enrichmeai.culvert:* on Maven Central — same contracts, Java owns the Beam/Dataflow execution layer; Python owns the Airflow runtime and dbt packaging.

Status

0.1.0 — first public release. Built and validated against a real GCP project (Cloud Run, BigQuery, Pub/Sub, event-driven end-to-end) before publishing. GCP adapters are production-shaped; the AWS family is Java-side; Azure is a roadmap skeleton. Honest limitations are documented per adapter in the source.

Docs, architecture, worked example deployments, and the engineering story: https://github.com/enrichmeai/culvert

Apache-2.0 licensed.