nx1-deployer image
v1.13.0New features
New features recently added to the NexusOne platform.AutoML
AutoML is a new NexusOne feature that makes building machine learning models simpler. You can now train and deploy three types of ML models directly from NexusOne: Each model also comes with built-in Shapley Additive Explanations support, so you can always see a clear explanation of why the model made a particular decision. AutoML comes with several new sub-features with their individual API endpoints.AutoML feature engineering
You can now clean and shape your data for training using AutoML feature engineering. It prepares your data into a form the model can learn from. The following endpoints let you create, approve, track, and retry that preparation before training starts:- Approve a proposal and start materialization
- Create a feature engineering proposal
- Get a feature engineering proposal
- Get materialization status
- List feature engineering jobs
- Re-run the crew on a failed job
AutoML training
You can now submit and manage machine learning model training jobs using AutoML training. The following endpoints let you list supported algorithms, submit, retrieve, or delete jobs, and manually trigger runs:- DAG status writeback with PSK auth
- Delete a training job
- Get a training job
- List supported AutoML algorithms
- List training jobs
- Manually trigger an AutoML training DAG
- Submit an AutoML training job
Deployments
You can now deploy a trained machine learning model to production as a batch job or an online endpoint. Batch jobs score data on a schedule. Online endpoints respond to requests in real time. The following endpoints let you create, list, promote, stop, and delete deployments:- Create a deployment for a model
- Get a deployment
- Get recent K8s events for an online deployment
- List a model’s deployments
- Get a prediction from an online deployment
- Replace data in an online deployment
- Send data to an online deployment
- Remove data from an online deployment
- Get options for an online deployment
- Get header metadata from an online deployment
- Update data in an online deployment
- Promote a deployment to a new model version or alias
- Run callback: batch DAG to API with PSK auth
- Set the Keycloak roles allowed to invoke or manage a deployment
- Soft-delete a deployment
- Stop a deployment
- Trigger an immediate batch run
Catalog
You can now search across all DataHub entity types from a single endpoint. Entity types are the categories of assets that DataHub tracks, such as datasets, charts, and dashboards.CrewAI
CrewAI comes with several new sub-features with their individual API endpoints.Crew events
You can now poll Crew execution events using the following endpoint:Crew runs
You can now view runs triggered by CrewAI using the following endpoints:Crew templates
You can now manage Crew templates using the following endpoints:DataHub proxy
You can now send requests directly to the DataHub General Metadata Service (GMS). GMS is the core backend that stores and manages all your metadata. The following endpoints let you read, write, and manage any DataHub entity by forwarding requests directly to GMS:- Get a DataHub GMS entity
- Update a DataHub GMS entity
- Create a DataHub GMS entity
- Delete a DataHub GMS entity
- Get DataHub GMS entity options
- Get header metadata from a DataHub GMS entity
- Patch a DataHub GMS entity
Documents
You can now store, search, and enrich documents in NexusOne. Documents are files or records indexed and queryable across your NexusOne environment. CrewAI can enrich the documents with metadata and push suggestions to DataHub. Server-Sent Events (SSE) streaming then lets you track that progress in real time. The following let you upload, retrieve, search, delete, and stream enrichment progress:- Confirm crew enrichment suggestions and push to DataHub
- Delete a document
- Fetch full document text by DataHub URN
- Get a document by ID
- List the caller’s documents
- Search DataHub documents
- Stream enrichment crew progress (SSE)
- Upload a document
MCP servers
You can now manage Model Context Protocol (MCP) server instances through the API. MCP is an open standard for connecting AI assistants to external tools and data sources. The following endpoints let you create, read, update, and delete MCP server instances:MLflow
MLflow gives you read-only access to the MLflow Model Registry, either through the UI or via API.API
You can now browse and inspect registered machine learning models and their versions directly using the MLflow API endpoints. The following endpoints let you list all registered models, fetch a specific model, and retrieve its full version history:- Get a registered MLflow model
- List all versions for a registered model
- List registered MLflow models
Observability
Each deployed NexusOne tenant now has 19 pre-built Grafana dashboards covering every NexusOne platform component. It also includes automated alerts that fire when components degrade or fail. Dashboards are available for the following NexusOne components:- AI API
- Airflow
- Apache Ranger
- Apache Spark
- DataHub
- Grafana
- Gravitino
- JupyterHub
- Keycloak
- Kyuubi
- Loki
- Metastore
- Mimir
- Namespace Overview
- PostgreSQL
- Superset
- Tempo
- Trino
- YuniKorn
Alerts fire after 5 minutes of a detected failure or degraded condition.
S3 gateway
NexusOne now includes an S3 gateway that allows you to programmatically access object storage through any S3-compatible tool using your existing NexusOne credentials instead of managing raw S3 keys. Keycloak authenticates every request, and Ranger authorizes them. This gives platform admins full control over who can access what.Bug fixes
Fixes to issues affecting apps or features on the NexusOne platform.OpenShift
Previously, platform components deployed on OpenShift clusters weren’t reachable at their expected URLs. NexusOne created routes without a host field, meaning components weren’t reachable at their expected URLs. NexusOne now assigns every component, such as JupyterHub, Trino, or Superset, a route with the correct hostname.Security patches
This release patches several CVEs across a few container images running in NexusOne.Enhancements
Enhancements to existing app features on the NexusOne platform.App manager
The App manager feature now has three new endpoints. Two let you attach and retrieve a JupyterHub notebook in an app pipeline. One retrieves a specific secret’s value for an app.- Attach a notebook reference to an app version
- Get a secret
- Get the notebook reference attached to an app version
Apache Ranger
The Ranger platform component is now a shared service across all tenants. Previously each NexusOne tenant ran its own Ranger instance, now it’s deployed once and shared across all tenants. Audit logs, the record of who accessed what data and when, were previously stored in Elasticsearch. They’re now stored in OpenSearch, which is NexusOne’s standard search backend.Ollama
The Ollama platform component is now a shared service across all deployed NexusOne tenants. Previously, each tenant ran its own Ollama instance, now it’s deployed once and shared across all tenants. This reduces resource overhead without any changes to the API or how you interact with it.Apache Spark
Spark jobs can now read from nested S3 directory structures automatically, without needing to explicitly list each subdirectory. This is particularly useful for partitioned datasets stored in S3.Upgrades
Version upgrades to existing apps on the NexusOne platform.Apache Ranger v2.8.0 upgrade
Apache Ranger now runs v2.8.0, upgraded from v2.6.2.
Apache Spark and JupyterHub
Apache Iceberg now runsv1.10.1. If you are running Spark jobs from JupyterHub, you get the latest Iceberg features
and bug fixes for table format operations.
Superset v6.1.0 upgrade
Superset now runs v6.1.0, upgraded from v4.1.2. This version includes new UI features, performance enhancements,
and bug fixes for dashboards and data exploration.