Managed SaaS releases¶
March SaaS feature announcements¶
March 2026
This page provides announcements of newly released features available in DataRobot's SaaS multi-tenant AI Platform, with links to additional resources. From the release center, you can also access past announcements and Self-Managed AI Platform release notes.
Agentic AI¶
PostgreSQL added as a vector database provider¶
You can now create a direct connection to PostgreSQL using the pgvector extension, which provides vector similarity search, ACID compliance, replication, point-in-time recovery, JOINs, and other PostgreSQL features. This is in addition to Pinecone, Elasticsearch, and Milvus for use as an external data connection for vector database creation.
New and retired LLMs¶
With this release, the following LLMs are retired and no longer available through the LLM gateway. As always, you can add an external integration to support specific organizational needs. See the availability page for a full list of supported LLMs.
- Arcee AI Virtuoso Large
- Arcee AI Coder-Large
- Arcee AI Maestro Reasoning
- Meta Llama 3 8B Instruct Lite
- Meta Llama 3.1 8B Instruct Turbo
- Meta Llama 3.2 3B Instruct Turbo
Registry Tools and MCP workflows¶
A new section has been added to Registry that allows users to manage agentic tools available to the deployment. Agentic tools provide a way for agents to interact with external systems, tools, and data sources.
The Registry > Tools page describes how to register and view and manage agentic tools for use with DataRobot's MCP server. For additional information on the MCP server, see the Model Context Protocol documentation.
Explore new GPU-optimized containers in the NIM Gallery¶
NVIDIA AI Enterprise and DataRobot provide a pre-built AI stack solution, designed to integrate with your organization's existing DataRobot infrastructure, which gives access to robust evaluation, governance, and monitoring features. This integration includes a comprehensive array of tools for end-to-end AI orchestration, accelerating your organization's data science pipelines to rapidly deploy production-grade AI applications on NVIDIA GPUs in DataRobot Serverless Compute.
In DataRobot, create custom AI applications tailored to your organization's needs by selecting NVIDIA Inference Microservices (NVIDIA NIM) from a gallery of AI applications and agents. NVIDIA NIM provides pre-built and pre-configured microservices within NVIDIA AI Enterprise, designed to accelerate the deployment of generative AI across enterprises.
With the release of version 11.7 in March 2026, DataRobot added new GPU-optimized containers to the NIM Gallery, including:
- Boltz-2
- cosmos-reason2-2b
- cosmos-reason2-8b
- diffdock
- nemotron-3-super-120b-a12b
- OpenFold3
New agent evaluation metrics¶
DataRobot now offers four additional metrics for evaluating agent performance. Three new operational metrics are available in the playground to provide insights into agent efficiency: agent latency measures the total time to execute an agent workflow including completions, tool calls, and moderation calculations; agent total tokens tracks token usage from LLM gateway calls or deployed LLMs with token count metrics enabled; and agent cost calculates the expense of calls to deployed LLMs when cost metrics are configured. These operational metrics leverage data from the OTel collector, which is configured by default in the agent templates. Additionally, a new quality metric for agent guideline adherence is available in Workshop, which uses an LLM as a judge to determine whether an agent's response follows a user-supplied guideline, returning true or false based on adherence.
For more information, see the documentation for Playground metrics and Workshop metrics.
NeMo Evaluator metrics in Workshop¶
Available as a private preview feature, NeMo Evaluator metrics are configurable in Workshop when assembling a custom agentic workflow or text generation model. On the Assemble tab, in the Evaluation and moderation section, click Configure to access the Configure evaluation and moderation panel. The NeMo metrics section contains the following new metrics:
| Evaluator metric | Description |
|---|---|
| Agent Goal Accuracy | Evaluate how well the agent fulfills the user's query. |
| Context Relevance | Measure how relevant the provided context is to the response. |
| Faithfulness | Evaluate whether the response stays faithful to the provided context using the NeMo Evaluator. |
| LLM Judge | Use a judge LLM to evaluate a user defined metric. |
| Response Groundedness | Evaluate whether the response is grounded in the provided context. |
| Response Relevancy | Measure how relevant the response is to the user's query. |
| Topic Adherence | Assess whether the response adheres to the expected topics. |
NeMo Evaluator metrics require a NeMo evaluator workload deployment, set in NeMo evaluator settings in the Configuration summary sidebar. Create the workload and workload deployment via the Workload API before you can select it; the Select a workload deployment dropdown shows "No options available" until a deployment exists.
For more information, see Configure evaluation and moderation.
Data¶
Databricks native connector now supports unstructured data¶
You can now ingest unstructured data from Databricks Volumes using the Databricks native connector. To connect to Databricks, go to Account Settings > Data connections or create a new vector database. Note that if you are already connected to the Databricks native connector, you must still create and configure a new connection to ingest unstructured data.
For more information, see the Databricks native connector reference documentation.
Support for Box added¶
Support for the Box connector has been added to NextGen in DataRobot. To connect to Box, go to Account Settings > Data connections or create a new vector database. Note that this connector only supports unstructured data, meaning you can only use it as a data source for vector databases.
For more information, see the Box reference documentation.
Jira, Confluence, and SharePoint are now supported OAuth providers¶
You can now add Jira, Confluence, and SharePoint as external OAuth providers in DataRobot. To add a new OAuth provider, go to Account settings > OAuth providers, and click + Add OAuth provider.
For more information, see the Manage OAuth providers.
Predictive AI¶
Observability configurations now compatible with Helm-native values¶
This release introduces improvements to observability by streamlining configuration to be compatible with Helm-native values. Instead of manually configuring multiple collectors, you now define your observability backends—like Splunk or Prometheus—once and assign them to a signal type: logs, metrics, or traces. This unified approach automatically handles the routing to the appropriate collectors, allowing you to use different backends for different signals without needing to understand the complex internal collector architecture. This reduces repetitive configuration and enables easier, more flexible telemetry export.
Applications¶
Applications moved to top-level navigation in DataRobot¶
You can now access all built applications, as well as access the Application Gallery, from the top-level navigation in DataRobot.
Previously, you had to go to the Applications tile in Registry to access both built applications and application sources. To create application sources or upload an application, you must still go to Registry > Application sources.
MLOps and predictions¶
Notification policies for individual custom jobs¶
On the Registry > Jobs tab, while configuring an individual custom job, the new Notifications tab provides the ability to add notification policies for custom jobs, using event triggers specific to custom jobs. To configure notifications for a job, click Create policy to add or define a policy for the job. You can use a policy template without changes or as the basis of a new policy with modifications. You can also create an entirely new notification policy.
In addition, the Notifications templates page in Console includes Custom job policy templates alongside deployment policy templates and channel templates, so you can author and maintain reusable notification policy templates for custom jobs separately from deployment policies.
For details on job-level configuration, template authoring, and channel behavior, see the Configure job notifications and Notification templates documentation.
Secure configuration exposure for models, jobs, and applications¶
A new organization-level feature setting enables the exposure of secure configuration values when they are shared with users and referenced by credentials used in runtime parameters. When this feature is enabled, secure configuration values are injected directly into the runtime parameters of a custom model, application, or job executed in the cluster. When this feature is disabled, the credential is injected using only the configuration ID, without exposing the underlying secure configuration values.
When the Enable Secure Config Exposure feature is active, the Share modal shows a warning that the secret might be exposed so administrators are aware that shared configs can be used in this way. Organization administrators control whether this feature is enabled.
Secret exposure in runtimes
Activating the Enable Secure Config Exposure feature flag causes secret values to be exposed in the container's runtime. When this feature is enabled for your organization, any credential created from a shared secure configuration and used as a runtime parameter in a custom model, application, or job exposes actual secret values (such as access keys and tokens) by injecting them into the runtime. Those secrets are then present in the container's runtime and can be accessed by custom code. Do not enable or use this capability unless you accept the risks inherent in exposing secrets in an uncontrolled container runtime. Use only when necessary and with appropriate governance.
For more information, see the Secure configuration exposure documentation.
Code first¶
Python client v3.14¶
Python client v3.14 is now generally available. For a complete list of changes introduced in v3.14, see the Python client changelog.
DataRobot REST API v2.43¶
DataRobot's v2.43 for the REST API is now generally available. For a complete list of changes introduced in v2.43, see the REST API changelog.
Deprecations and migrations¶
Upcoming deprecation of Aryn engine¶
With the next release, DataRobot is deprecating Aryn’s optical character recognition (OCR) API for self-managed, multi-tenant, and single-tenant deployments.
All product and company names are trademarks™ or registered® trademarks of their respective holders. Use of them does not imply any affiliation with or endorsement by them.