hopsworks-api

Use when writing and deploying an interactive agent (e.g. a LlamaIndex program) as a served Hopsworks deployment.

Use when writing Streamlit or custom Python apps for Hopsworks, deploying from HopsFS or Git, migrating legacy apps off `APP_BASE_URL_PATH`, or managing app routing/readiness, monitoring, and public sharing. Auto-invoke when user wants to create a dashboard, deploy a Python app to Hopsworks, configure app routing/readiness, or access the feature store from an app. Input an app source + env + memory; output a running app and its URL.

Mount or ingest a table from a supported datasource. Mount tables from a datasource as an external feature group or ingest data into a new feature group using DLTHub. Auto-invoke when user works with external data (Snowflake, BigQuery, Redshift, S3, ADLS, GCS, JDBC, SQL, Databricks Unity Catalog, Postgres, MySQL, Oracle, SAP, MongoDB, CRM, REST APIs).

Use when writing Python code that creates, inserts into, or manages tables or feature groups. Auto-invoke when user writes feature pipelines, feature engineering code, or asks about feature group best practices (online vs offline, batching, OOM, materialization, embeddings, statistics).

Use when a Hopsworks job, app, or deployment needs Python libraries that are not in a base environment. Clone a base environment and install requirements or a wheel into the clone. Auto-invoke when the user hits a missing-package error in a job/app/deployment, asks to install custom dependencies, add a pip requirement, install a wheel, or pick which Python environment a workload should run in.

Use when the user asks where something lives in the Hopsworks UI or how to reach a page (Data Sources, Feature Groups, Feature Views, Model Registry, Deployments, Apps, Agents, Jobs, Jupyter, Project Settings). Knowledge skill: the project-scoped sidebar layout and what sits under each section, so you can point users to the right place.

Use when working with Hopsworks — feature groups, feature views, training datasets, storage connectors, models, deployments, projects, jobs, and datasets. Auto-invoke when the user discusses feature engineering, feature store operations, ML pipelines, model serving, external data sources, Superset, or needs to interact with Hopsworks.

Use when training an ML model. Load training data from an existing feature view, train a model with an appropriate ML framework, evaluate it, and register the model and its evaluation (metrics, plots) in the Hopsworks model registry. Input - a feature view; Output - a registered model.

Use when writing Python code that creates, queries, or manages Hopsworks feature views via the hsfs SDK. Auto-invoke when user builds feature views, selects features, applies transformations, creates training data, retrieves feature vectors, or asks about feature view best practices (labels, filters, joins, transformations, online serving, embeddings).

Use when writing Python code to monitor a Hopsworks feature store with descriptive statistics, feature/drift monitoring, Great Expectations data validation, alerts, and inference feature logging. Auto-invoke when the user wants to monitor feature drift, set up data quality checks, validate data on ingestion, configure alerts, compute statistics, or use Great Expectations with Hopsworks.

Use when writing code for model deployment, online inference, predictor scripts, or on-demand transformations in Hopsworks. Auto-invoke when user wants to deploy models, write predictor.py files, retrieve precomputed features for serving, create on-demand transformation functions, or configure model serving. Input registered model + online feature view; output a running KServe endpoint.

Use when building Superset charts or dashboards inside Hopsworks via the Python SDK. Auto-invoke when the user wants to create Superset charts/dashboards/datasets, visualize a feature group in Superset, or interact with `project.get_superset_api()`. Input an offline-materialized FG; output a published Superset dashboard + URL.

Create and schedule/run a feature pipeline program from additional user input. Build on the ML system requirements.

Find available data in Hopsworks from feature groups, data sources, free-text search, and files.

Use whenever you write PySpark code in a Hopsworks terminal (terminal-spark) or any project that runs PySpark. Auto-invoke when the user writes `spark` or `pyspark`.

Query Hopsworks feature store tables via Trino SQL using the hops CLI.

Parses unstructured data from files and writes it to a feature group.

Use when creating, configuring, scheduling, or running claude-code or codex jobs/workflows.

Use when writing Python or PySpark code for batch inference with Hopsworks. Auto-invoke when user wants to retrieve batch data from feature views, use spine groups for point-in-time joins, download models from the model registry for batch prediction, or build batch scoring pipelines. Input feature view + registered model → output predictions, logged or persisted.

Use when creating, configuring, scheduling, or running Hopsworks jobs or Airflow jobs/DAGs/workflows. Input a script in HopsFS plus job config; output a created/scheduled job and its executions.

Use when writing Python code for Hopsworks transformation functions, including built-in transforms, custom @udf transforms, model-dependent (statistics-based) transforms, on-demand (request-time) transforms, and the transformation store. Auto-invoke when the user applies scaling/encoding/imputation to features, writes a @udf, creates on-demand features from request parameters, registers a transformation function, or asks how transformation renaming/execution modes work.

The dimensions to cover when doing EDA on feature-view training data before training a model in Hopsworks, covering dataset profiling, target analysis, per-feature analysis, and leakage detection. Force into context when planning or reviewing EDA, deciding what to profile, or checking for label leakage and train/test contamination. Reference companion to the hops-eda action skill.

Exploratory Data Analysis before training an ML model. Auto-invoke when the user wants to profile feature-view training data, check for leakage, or produce an EDA report before building a feature view or training.

Create a specification for an ML system as a composition of datasources and separate feature, training, and inference (FTI) pipelines (batch or online). There will be an order relationship between the pipelines, some will be blocked-by other pipelines. Write down the specification as a markdown file in reqs/. Ask the user if they want to then implement the system.

Use when working with Hopsworks — feature groups, feature views, training datasets, storage connectors, models, deployments, projects, jobs, and datasets. Auto-invoke when the user discusses feature engineering, feature store operations, ML pipelines, model serving, external data sources, Superset, or needs to interact with Hopsworks.