Release notes

September 2023

  • Verta MLflow Integration ALPHA: Bulk import MLflow models to Verta Model Catalog

    • Verta Model Catalog helps you create a unified model inventory, regardless of where the model was originally built or trained.

    • To simplify the process of importing models into Verta Model Catalog, we have developed a Verta MLflow integration that allows you to batch import MLflow models with just a few clicks.

    • To learn more follow this guide.

  • GPU support

    • Verta now supports the ability to run endpoints on GPUs.

    • Please contact if you need access to this feature.

    • To learn more click here.

  • Custom attribute categories

    • Custom attributes are crucial for companies to develop their Enterprise Ontology in Catalog.

    • We now offer the option to create custom attribute categories to enhance their usefulness across different company departments.

    • Custom attributes are grouped together based on these categories, simplifying management and access.

    • To learn more click here.

  • Custom attribute source

    • Custom attributes can now be configured to track the ingest source, last updated user-ID, and timestamp information.

    • To learn more click here.

  • New Govern tab for release checklist

    • Release checklists are valuable for policy management throughout a model's lifecycle, which is why we've introduced a dedicated "Govern" tab for Model Versions to monitor all checklists and their statuses.

    • You have the option to filter checklists by custom attributes or release stages and perform actions on active checklists.

    • To learn more click here.

Enhancements and Resolved Issues

  • Model versions are now accessible from the Model Summary page.

  • In the checklist description and evidence sections, links are displayed as clickable hyperlinks.

July 2023

  • Introducing AI Assisted Model Documentation BETA

    • By using AI Assisted Documentation data scientists will be able to write more detailed and accurate model documentation in less time.

    • AI Assisted Documentation works by leveraging a large-language-model (LLM) provided by OpenAI. Verta has fine tuned a selection of “prompts” that will return an ideal response from the LLM.

    • To learn more follow this guide.

  • Model schema specification and validation with Open API and Pydantic

    • Leveraging the new model schema specification feature, you can now specify a schema for the input and output data of your deployed model.

    • In addition, you can optionally enforce validation of input and output when predictions are made against the model.

    • Model Schema uses OpenAPI standards and integrates seamlessly with Pydantic, which makes it easy to create schemas and correlate JSON objects.

    • To learn more follow this guide.

  • Manage custom REST headers during predictions

    • Verta now offers the functionality to include custom REST headers in the model prediction service. For instance, users can pass access tokens for connecting with third-party services by adding new parameters as headers to their model's predict function. When the predict function is called, all request headers will be automatically populated into those parameters.

    • To learn more click here.

Enhancements and Resolved Issues

  • New sign-up page: Verta now has a new and improved sign-up page.

June 2023

New Features

  • New Documentation Text Editor

    • Our Model Catalog now includes a new model documentation text editor, facilitating the capture of model documentation and change logs for models and model versions respectively.

    • Download github-ready markdown files to keep your systems in sync.

  • Private Ubuntu APT integration

    • Verta platform now provides support for integrating with private APT repositories, allowing companies to host their Debian packages.

    • System administrators can configure one or multiple APT repositories. When models are deployed in Verta, the platform scans these configured repositories for dependencies instead of relying on a public repository.

    • To learn more click here.

    • If you require access to this feature, please inform your Verta support team.

  • Tag PII models to prevent model input logging

    • In the Verta Model Catalog, a new PII tag is available for registered models. It can be toggled on/off using the web app and client SDK.

    • When the PII tag is toggled on, the system captures the PII configuration during endpoint deployment, ensuring no logging of input features for that model.

    • To learn more click here.

  • Display release checklists based on custom attributes

    • Organization admins can now configure different checklists for different models, based on custom attributes. For example - set 3 different checklists based on your custom 3-tier risk ontology. As models are marked by risk level, see the correct release checklist.

Enhancements and Resolved Issues

  • We have expanded custom attributes to support short text and numeric fields. It will help admins to customize catalog and create their own enterprise ontology.

  • The model endpoint summary page has been refreshed, improving the accessibility of key information.

  • For customers using external vulnerability scanning feature, the model scan button in Catalog has been updated to allow scanning when there is an existing/latest build. With this change a new build does not get triggered when user clicks scan thus eliminating undesirable build creation.

  • Invite team members to join your catalog or collaborate with the ML team by email from the homepage or any model.

May 2023

New Features

  • Making Verta Model Catalog configurable with custom attributes

    • Admins can now configure custom fields in Verta Model Catalog, allowing them to create their own enterprise ontology.

    • This feature is beneficial for setting model risk levels, assigning owners from product, QA, IT, and Risk departments, assigning business units, and defining other custom model attributes.

    • Custom attributes can be configured for both models and model versions.

    • To learn more follow this guide.

  • Kafka integration BETA

    • Verta endpoints now support near real-time/asynchronous models with Kafka integration.

    • System admins can easily connect their Kafka streaming servers to Verta for self-service integration.

    • Once integrated, Verta endpoints can read from a Kafka topic, perform predictions, and write output to a Kafka topic.

    • To learn more follow this guide.

    • If you require access to this feature, please inform your Verta support team.

  • Executive dashboard ALPHA

    • New executive dashboard view available for AI/ML model metrics, including model health, compliance, and velocity.

    • Metrics include production model count, models by stage or refresh age, and prediction endpoint count over time.

    • Note: The feature is currently available only for SaaS and VPC customers.

Enhancements and Resolved Issues

  • Introducing a new self-guided onboarding experience in Verta that assists users in effortlessly cataloging their models and accessing relevant examples and documentation.

  • The Verta UI now includes a user-friendly drag and drop document upload feature, allowing governance team members and non-technical stakeholders to easily manage artifacts such as Model Risk Documents, Requirements documents, and more within the model catalog.

  • The client APIs have been enhanced to facilitate the automation of model promotions. The following improvements have been made:

    • List builds and scan status for a specific model version.

    • List the latest build and scan status for an endpoint.

    • Trigger a scan for a particular build.

    • Please note that these updated build scan APIs are specifically available for the external scanning feature.

April 2023

New Features

  • Introducing Batch-Based Processing Deployments

    • Model versions may now be registered with batch processing configuration. This will allow for Verta endpoints that expect to receive large data frames to process in batches

    • The batch predict feature allows you to make predictions against a large Pandas DataFrame. The client's batch_predict() method will split the DataFrame into batches of the specified size for prediction requests and then collect the predictions to return as a single DataFrame

    • To learn more follow this guide

  • Release Checklists enhancements

    • Items on a release checklist have optional access controls - organization admins can configure checklist items so that only the approved user group may complete it

    • Release checklists can prevent transition to a specific stage. Organization admins can assign a checklist to apply to a stage transition and users will be prevented from starting the transition until the corresponding checklist is complete

  • Validate a model's dependencies locally and speed up iteration cycle

    • Verta now provides convenience functions for scanning a provided model class for dependencies and identifying any missing packages in the provided environment

    • This can help catch problems early in the development process, and avoid time wasted debugging deployment issues

Enhancements and Resolved Issues

  • Verta conducted our standard monthly vulnerability detection and remediation process

March 2023

New Features

  • Introducing Release Readiness & Governance Checklists

    • Create and customize checklists for your team to review and complete as model versions advance through the lifecycle

    • Improve the safety and governance of model releases without slowing them down with heavy-handed processes

    • Generates auditable paper trails

    • To get started customizing your own checklists, upgrade to the new permission system and follow along with this guide

  • Speed up iteration on models for deployment

    • Client based tools to verify missed dependencies, external libraries or data type conversion issues before registering a model

    • Provide the ability to add init test client function that runs at the end of the build to automatically test an endpoint and surface results - prevents deploying erroneous endpoints

  • New Permission and Access System

    • Simplifies access management for data scientists and users and provides more granular control for organization administrators

    • Straightforward connection to Enterprise IdP systems like AD and Okta

    • Coordinate a migration for your private environment with your Verta contact

Enhancements and Resolved Issues

  • Verta conducted our standard monthly vulnerability detection and remediation process

  • Verta supports bulk ingestion of models to Catalog from ML Flow. Find the script you need in our examples repo

February 2023

New Features

  • Multiple Namespace Support is now in beta

    • Use Kubernetes namespaces to isolate Verta endpoints to different nodes in your cluster

    • Isolate Business units and give them different resource quotas

  • Log all model traffic with Model Data Logging

    • All model traffic, including the input data sent for inference, the results of the inference and any intermediary data, can be stored now

    • Store and search your model traffic with Amazon Athena to trace model performance issues and results

  • CloudWatch Integration

    • Verta now integrates with CloudWatch for APM monitoring

    • Contact us to get started with the CloudWatch configuration if your team needs it

  • Dedicated resource mode

    • You can now configure dedicated machines for high throughput and compute heavy endpoints.

    • Dedicated resource mode blocks entire nodes for model inferences of an endpoint and prevents noisy neighbor problems.

Enhancements and Resolved Issues

  • There is a new platform login page with updated visual

  • Track all deploys of your models on the RMV Release page, now including listing cloud provider service or naming edge devices where the model is running

  • Pending approval (of model version stage changes) will be collated for all versions on a new tab for the model titled “Pending Approval”. Quickly review and act on all requests

  • The Model Versions tab will now support a card and a list view. From list view you can select multiple model versions to compare details across versions

  • On model versions, the release readiness checklist has been updated to allow users to manually “check off” (complete) items, and to log who completed the item and the timestamp of the action

  • Custom prediction id support: With the deployed_model.predict_with_id function, pass your custom UUID to track model inferences

  • Verta conducted our standard monthly vulnerability detection and remediation process

  • An issue with SAML login was resolved

  • Improvements to resolve the sporadic long latency issue and sporadic 5XX related to Load Balances and improper draining and removal of nodes in AWS

  • Other performance improvements related to web UI and authentication systems, including validators to prevent issues with variable names and canary parameters for deployment

November/December 2022

New Features

  • Model Catalog Integrate and Reproduce Pages

    • Two new tabs on each model version capture the information needed to test and reproduce models and integrate them with existing systems.

    • The Reproduce tab shows information captured when the model is registered that lets collaborators understand and models lineage and reproduce the model. The model’s environment, experiments that were run upstream, code version, datasets and any attributes logged on the model are visible.

    • The Integrate page collates the necessary details for any engineering partner who is looking to integrate the model into existing tools. Artifacts and and the models API (as JSON and easily readable widget) are available

    • These new pages replace the model packages, api and insights tabs on model versions. No features have been removed.

  • Model Catalog Log External Deploys

    • Document and record where your models are deployed, no matter if they’re in Verta, another cloud provider or on an edge devices

    • Collaborators can note the location and description of all deploy locations (directly beneath where they download the model container on the release page) so all locations where the model is running are documented

    • Verta endpoints are automatically documented

Enhancements and Resolved Issues

  • The model monitoring page has been updated to include a newer, easier to use and smaller filtering tool.

  • Model Version Stage Changes (moving between dev, staging and production) has been updated to allow for more possible stage changes as long as the requisite approvals are granted. If users are reverting or skipping suggested changes, a notification will appear so they are aware.

  • Stage changes for a model will now be recorded as 3 steps on the activity log, a request, an approval and a commit. The commit is assigned to the original requestor and occurs automatically once the final approval is given.

  • Verta conducted our standard monthly vulnerability detection and remediation process

  • Our Vulnerability Scanning feature has moved to the Release page on a model version, and the tool has been updated to be easier to read.

  • The Activity Log is now on the Registered Model level and shows activity for all versions.

  • Platform audit logs can now log Model API/inference activity, including in a separate location from platform operational logs.

October 2022

New Features

  • Model Landing Page is now Model Cards

    • Information about models has been reorganized so all team members can quickly understand key details about each model in your organization’s catalog

    • This includes a control panel to track each version’s progress toward production, an easy-to-use view of the model’s API with quick descriptions about what data it ingests and produces, and usability improvements for model promotion, model locking, and reviewing the model activity log

    • Quick links help you find the endpoints and monitors for all versions of the model

  • Model Release Management Page

    • A new release page on registered model versions helps you create and track releases of your model

    • All endpoints (along with owners, labels, and status) are listed with a simplified way to create new endpoints directly from the model version

    • Download the runnable Docker container for the model directly from the release page

  • H2O Model Support

    • You can now easily integrate H2O models with the Verta platform

    • See our documentation for an example of how to register, deploy, and run H2O models

  • Monitoring Alert History

    • You can now access a detailed history of an alert rule in model monitoring

    • Alert history gives a log of all changes to the alert, organized by timestamp, that includes any updates to alert conditions and any changes to alert status

Other Enhancements and Resolved Issues

  • Health Check Endpoints are now available for basic application monitoring for on-premise customers - Learn More

  • A model’s activity can now be collated across all versions, filtered and exported as a CSV

  • We have added a new “No data” alert status in model monitoring; if the alert rule metric has not received any inferences during the alert evaluation period the status of the alert rule is shown as "No data"

  • You can now easily specify the type of the model (Classification or Regression) using the field task_type in Python client or via WebUI in the model catalog; monitoring dashboards will pre-compute the right set of metrics and dashboards based on the model task type

  • Verta has added support for Python 3.10

  • Verta no longer supports Python 2.7, 3.5, and 3.6

  • Verta conducted our standard monthly vulnerability detection and remediation process

  • An issue preventing endpoints from being removed was identified and resolved

September 2022

New Features

  • Model Catalog on Homepage

    • The platform homepage now includes a Catalog widget that lets you get quick operational insight on the models being registered, updated and maintained on your team.

    • Get at-a-glance insight into the most recently updated model versions and the numbers of models in production or in other stages.

  • Model Catalog New Filters and Model Metadata

    • Faster and better model inventory management with new filters and metadata attributes helps your team reuse and get more value from existing models

    • Look up models by owners, if they have live endpoints and by machine learning task

    • Assign data type and task type to models (ie: “text classification”, “audio transcription”) via the UI or Python client to categorize your catalog

  • New Model Documentation Template

    • The Verta platform now provides a robust Model Documentation template based on model documentation research from Google and IBM including prompts about model performance, limitations, risks, dependencies and opportunities.

    • Enable your teams to easily follow model documentation best practices in Verta. Quickly embed images, links, and emoji so key data and facts can be easily highlighted and shared

    • Transition projects between data scientists with ease and allows new team members to get up to speed on models quickly

  • Custom Vulnerability Scanning Webhook

    • Customize your model security scanning with a quick integration with the Verta platform.

    • One-click vulnerability scanning for data scientists so they can quickly check all models before release

    • Receive a webhook with build information to scan, set up custom business rules and return a pass or fail result to Verta’s platform to give quick feedback to the data science teams

Other Enhancements and Resolved Issues

  • Model Service will no longer use Flask, and instead will leverage CherryPy. Older model endpoints running Flask will be supported and new model endpoings will leverage CherryPy.

  • We conducted our standard monthly vulnerability detection and remediation process

  • System Admin features for Verta customers have been hardened and system admins can now act as full users within the platform

  • Verta no longer supports Python 2.7 and 3.5, which have been retired

  • Deleting models via the UI now requires a long-press confirmation to reduce risks of outages or issues

  • Navigation has been slightly updated to feature a new tab system

  • Headers for Prediction Responses will now include endpoint owner information

August 2022

New Features

  • Updates to Platform Navigation: Our August release includes exciting UI enhancements featuring updated navigation. This update makes our platform more accessible and leaves more room for pages with a narrower, sleeker set of navigational tools. Here’s a few changes to be aware of:

    • Experiment Tracking can now be found under the Experiments (flask icon) and Data (database icon) tabs.

    • Endpoints has been renamed and can be on the Deployment tab (rocket icon)

    • Registry has been renamed Catalog and can be found under the Catalog tab (card icon)

    • Your profile and admin pages have been moved to the bottom left

    • You can now see a lifesaver icon that links to all help features in the platform

    • Workspace and Search remain in the top right corner

    • If you have any questions or feedback please let us know at

  • Welcome Home to Homepage: With this release we’re introducing the platform Homepage, will will be your new landing page whenever you sign in:

    • For new users, homepage will show a Welcome widget (that they can dismiss) with helpful introductory steps and videos. You can always recall the Get Started widget under the new help lifesaver icon

    • Returning users will be shown a widget that summarizes the models currently being monitored for your current workspace, included summary stats on model traffic and alerts in the last 30 days, and a direct link to the models getting the most traffic this week.

    • We have more widgets coming soon, and if you have any questions or requests please contact

  • Introducing Model Catalog: Model Registry is now Model Catalog - your organization’s unified repository of models in development and production:

    • All features of our model registry are available, and we’ve updated the landing page to be much friendlier to users of all types

    • Catalog lets you organize and attach models, model metadata, documentation, implementation details and compliance details in one place

    • Searchable, filterable and sortable so everyone can easily find live, in development and legacy models with no more shoulder-tapping of the ML or operations team

    • Increase the value of existing models by making them more accessible and reusable and reduce operational overhead spent compiling and documenting information about models

Other Enhancements and Resolved Issues

  • Automatic Alert Creation for Model Monitoring is now available, and alerts will be created as soon as you set up monitoring on an endpoint

  • There is a new event on the platform event system, the Promotion to Production event which allows you to set up automations when a model version is moved to production. Please contact if you need access to this webhook.

  • We conducted the standard monthly vulnerability detection and remediation process

  • Admins may now configure Verta On Premises to set all model builds to generate fully self-contained images

July 2022

New Features

  • Monitoring Alerts: Automatically detect drift or model performance degradations with configurable alerting rules and get notified. Users can easily modify and/or create new rules. The platform will notify users when an alert threshold is crossed. Learn more: here

  • Monitor Regression Models: Model Monitoring in the Verta platform now supports an additional model type - regression models. This includes drift and outlier alerting for regression models. The platform will automatically compute performance metrics such as R-square, RMSE, MSE and MAE. Learn more: here

  • Support for SCIM provisioning: Customers can now integrate the Verta platform with their chosen SCIM provider (such as Active Directory or OKTA) to automatically provision, update and deactivate users and groups. Learn more: here

Other Enhancements and Resolved Issues

  • Monitoring - Customers can now set granular permissions on model monitors, including full RBAC system support

  • UI Enhancements - The platform will show many updated components to improve ease of use. The settings pages for sharing and visibility have been completely updated.

  • System Monitoring - The Verta platform now monitors model deployment and provides our Verta support engineers better insight into system instability.

  • Webhook Event Service ALPHA release - We are introducing a webhook event system to the Verta platform to allow for easier custom integrations. If you’d like to learn more or give feedback on the ALPHA please reach out to

  • Vulnerability Remediation - We have a monthly security vulnerability identification and resolution process, which was followed this month.

  • Automated logging - The system powering automated audit logging has been updated to remove the need for Elasticsearch. This release also introduces a new folder structure to allow for secure isolation of different namespaces and easier searching of the audit logs.

Bug fixes

  • System admins can delete entities they own

  • External package configurations will not be deleted upon upgrade of dedicated Verta install

Last updated