DataHub Plugin

DATAHUB
WRITTEN AS IT RUNS

You already run DataHub. Strongly fills it. Every workflow run, model, AutoML job, and drift result lands in your catalog as it happens. No exports. No nightly job. No diagram to redraw after the work changes.

Real time
Sync on every run
19
Native platforms
0
Manual exports

Your AI Metadata,
In the Catalog You Trust

Install the plugin, point it at your DataHub, and the metadata your workflows and models produce shows up where your data team already looks.

Lineage On Every Run

Each workflow becomes a dataFlow. Each node becomes a dataJob. Each execution becomes a run instance with status and duration. The graph in your catalog is the graph that ran.

dataFlow dataJob Run history

Model and AutoML Sync

Registered models land as mlModel entities with training metrics, hyperparameters, and training-dataset lineage. AutoML jobs carry their leaderboard and winning model.

mlModel Metrics Leaderboards

Drift As Assertions

Drift results post as DataHub assertions with a pass or fail per run. A model going out of bounds surfaces in the same view your team checks for data-quality breaks.

Assertions Pass / fail Drift score

Native Platform Mapping

A node that reads Snowflake maps to a Snowflake dataset. Postgres to Postgres. S3 to S3. The same URNs DataHub uses, so AI lineage connects to assets your other tools ingested.

Snowflake BigQuery S3

Installs As A Plugin

No pod to provision, no service to run. Point it at your DataHub GMS with a personal access token, a DataHub Cloud token, or no auth for a local instance. Enable or disable per feature.

No extra infra PAT or Cloud Toggle per feature

SDK For The Rest

Uploaded a file and trained on it? Register a dataset or model the platform did not produce through the SDK. It uses the same governed path as the automatic hooks, so no extra credentials.

emit_dataset emit_model Governed
5
Lifecycle Events
6
Entity Types
19
Native Platforms
Upsert
Idempotent By URN

Strongly Concepts to DataHub Entities

Every emission is an idempotent upsert keyed by URN, so re-running a workflow updates the catalog in place instead of duplicating it.

Workflow Lineage

The structure and the run history of every workflow, mapped the way DataHub models orchestration. Node-to-node edges come from the execution span tree.

dataFlow dataJob dataProcessInstance
What lands:
  • Workflow to dataFlow, with a link back to the builder
  • Each node to a dataJob, with type and node id
  • Each run to a process instance, with status and duration
  • Source and addon reads mapped to native-platform datasets

Models, AutoML, and Drift

The model registry, AutoML runs, and drift monitors, mapped to the ML and assertion entities your governance views already understand.

mlModel dataset assertion
What lands:
  • Registered models with metrics and hyperparameters
  • Training datasets linked as upstream lineage
  • AutoML leaderboard and winning model
  • Drift runs as assertion results, pass or fail

Lineage That Connects

Where the source is known, Strongly names the dataset under its real platform. Anything without a native home gets a first-class Strongly platform. Nothing is guessed.

Snowflake BigQuery Postgres MySQL Redshift S3 MongoDB Databricks Kafka Elasticsearch ClickHouse Trino Oracle SQL Server
# Register an uploaded dataset and the model trained on it
client.datahub.emit_dataset(name="q3-claims.csv", platform="s3")
client.datahub.emit_model(name="claims-triage", dataset_name="q3-claims.csv")

Keep Your Catalog
Current By Default

See the DataHub plugin emit a live workflow run, a model, and a drift result into a catalog of your own. Bring your GMS URL and we will walk it end to end.