Key Features

Folder Deletion in Data Center

Data Center now supports deleting folders directly, with behavior tailored to the folder version underneath. V1 folders perform a soft delete with progress feedback so you can track the operation as it runs. V2 folders perform a hard delete, and if a folder still contains items, we block the delete and explain why rather than silently failing or removing content you didn't intend to lose.

App-Agnostic Deep Linking

Links into aiWARE can now open a specific Center or Panel directly, regardless of which application is hosting them. This makes it possible to share a link to an exact view, such as a particular Processing Center job or Data Center folder, and have it land the recipient there directly instead of at a generic app landing page.

Translation Viewer

We've added a Translation Viewer to the Data Detail experience, giving translated text its own dedicated view alongside the other engine result viewers rather than requiring users to dig through raw engine output.

Geographic Search for Media

searchMedia now supports geographic search, including geo_bounding_box filtering and geohash_grid aggregation. This enables map-based filtering of large result sets, so users working with geotagged media can narrow results to a region on a map instead of only filtering by text or metadata fields.

HLS Multi-TDO Stream Stitching

aiWARE can now generate HLS streaming manifests that stitch together multiple TDOs into a single continuous stream. This is useful for scenarios where a logical recording spans several underlying TDOs and needs to be played back as one uninterrupted stream.

Automate Studio Node-RED Module Catalogue

Automate Studio now includes a custom Node-RED module catalogue, dynamically populated from published @veritone/node-red-contrib npm packages. This makes it easier to discover and add Veritone-maintained Node-RED nodes to a flow without leaving Studio.

Automatic DNS for Vanity Domains

Custom (vanity) domains configured for an application now get their DNS records created automatically, so they route correctly to the application's load balancer without a manual DNS setup step.

aiWare Updates

aiWare UI Updates

• Fixed an issue where desktop viewers did not send the required origin header when fetching signed asset redirect URIs, causing documents to fail to load due to CORS errors.

• Fixed a CORS configuration issue that prevented viewers from loading media within Data Center after a server migration.

• Fixed a startup crash in the CMS application caused by a compromised transitive dependency that prevented the service from starting after deployment.

• Fixed an issue where creating a new folder in Data Center failed to apply group permission sets due to an invalid resource type value being sent to the setACLs mutation.

• Fixed an issue where clicking Deploy or Save Draft in Automate would spin indefinitely without completing, caused by malformed request URLs.

• Fixed an issue where individual facial images could not be deleted from an Entity in Resource Center without deleting the entire Entity.

• Fixed an issue where duplicate Library names were not blocked when creating a new Library in Resource Center.

• Fixed an issue where enabling the Ratings and Nielsen toggles in Admin Center did not make Nielsen data appear in the Discovery UI.

• Fixed an issue where opening a transcript search result link did not automatically scroll viewers to the matching location.

• Updated login region selection so users are no longer defaulted to CA when signing in from another region.

• Updated the OCR timeline viewer to group repeated text results and handle confidence scores, bounding boxes, and long text consistently with other engine result types.

• Added a Scheduled Job Name search category to the Data Center search bar.

• Added a config-override indicator to the Edge Admin UI's engine list so engines with resource overrides can be identified at a glance.

• Added the ability to configure per-engine CPU and memory resource request and limit values from the Edge Admin UI.

• Restored the Add Packages button in Admin Center and Instance Admin Center for managing an organization's packages.

• Updated the viewers app to load its Content-Security-Policy from edge-controller configuration.

aiWare API Updates

• Fixed an issue where special characters such as parentheses in OCI asset keys blocked retrieval of the associated assets.

• Fixed an issue where incorrectly constructed object storage keys could leave some assets inaccessible or fail to promote between storage tiers.

• Fixed an issue where signed URLs for certain storage buckets in an external AWS account incorrectly used Oracle Cloud credentials instead of the AWS instance profile, causing access errors.

• Fixed an issue where the engineResults GraphQL query could return an internal server error due to a circular data reference, causing media playback and engine-results panels to fail to load.

• Fixed an issue where reverting OCI storage settings caused signed URI requests for existing assets to throw internal server errors instead of returning an unsigned URI.

• Fixed an issue where the built-in GraphiQL interface failed to render after a dependency upgrade.

• Fixed an issue where listing engines by ID could time out due to an inefficient database query on large ID filter sets.

• Fixed an issue where unbounded Elasticsearch connections could exhaust the search thread pool and cause search requests to time out.

• Fixed an authorization check on getApplicationLogo that caused the app switcher to show generic default icons for users without Developer, CMS, Discovery, or Org Admin access.

• Fixed performance and reliability issues in the mdpAddSegmentsBulk GraphQL mutation to handle high-volume segment additions without exhausting database connections.

• Fixed the pipeline that indexes engine GPS/geolocation output so location data reliably reaches the search index for map-based filtering.

• Fixed service token permissions not being persisted to configuration, which could result in newly created tokens having no assigned rights.

• Fixed stale service host and path prefixes left over after an ingress rework that could produce incorrect service URLs.

• Fixed LLM Gateway logging that was silently dropping logs after a litellm version bump, and made the log builder compatible with both versions.

• Fixed invite email links so they route directly to the correct login domain instead of through an intermediate apex record.

• Fixed SCIM provisioning so re-provisioning a previously deactivated user reactivates the existing account instead of failing.

• Fixed the virtual asset HEAD endpoint on OCI so it returns correct object metadata instead of a signature-mismatch error, and added a Veritone-Virtual-Asset response header to identify virtual-asset responses.

• Fixed deleted engine configuration overrides not refreshing the in-memory cache and requiring an exact value match to delete.

• Fixed inconsistent values returned by Edge Admin config GET endpoints by adding an opt-in refresh parameter that forces a fresh read instead of a stale per-pod cache.

• Added the filename as the last segment of the virtual asset URL path so processing engines using GetChunkedUrl can save files by name.

• Added a redirect=false parameter to virtual asset URL requests so applications can retrieve the source URI directly instead of following a redirect.

• Added support for including Libraries as a resource in Packages so they can be distributed and shared across organizations.

• Added an insert-only configuration API so importing engine configuration no longer overwrites manually configured values.

• Added a DELETE /admin/config/{ConfigSection}/{ConfigKey} endpoint allowing per-engine configuration overrides to be cleared back to the default.

• Updated several GraphQL resolvers to return stateless virtual asset URIs instead of signed URLs for image, icon, and logo fields.

• Updated LLM Gateway logging to retain request metadata using a static placeholder instead of raw text when redaction is unavailable, preventing PII from being stored.

• Updated the veritone-sdk file upload flow to verify the uploaded file is accessible before creating its TDO and asset, reducing S3 eventual-consistency race conditions.

• Added audit log events for object-level permission grants, revocations, policy updates, and access checks.

• Various security improvements.

aiWare Processing Updates

• Fixed an issue where a customer's original root folder could go missing, hiding previously uploaded files and folders from the UI even though the underlying assets remained intact.

• Fixed an issue in the Engine Toolkit where engine output chunks could be silently dropped, causing tasks to report success despite missing results.

• Fixed an issue where task status could regress or diverge from its actual state on multi-instance chunk-engine jobs, causing jobs to appear stuck as running or lose completed results.

• Fixed a crash in the engine SDK that could silently drop a request when creating a schema definition object for an engine without a published output schema.

• Fixed an issue where live HLS streams could be incorrectly terminated with an end-of-stream marker due to mishandling of the requestLive flag.

• Fixed job cancellation to also stop actively running work and drain the associated engine instances instead of leaving them running indefinitely.

• Fixed a bug where a per-engine configuration override could suppress the global default for other engines, silently disabling automatic resource resizing.

• Improved S3-ingested asset and file names to use a meaningful base name plus context (such as Camera, Photos, or app name) instead of the full source URI.

• Improved URI-based file discovery to capture file size and modification date when available.

• Removed a SQLite write bottleneck in the Track processing pipeline by tuning database durability settings and allowing the database to run from a faster location.

• Added a circuit breaker for S3 Ingest SQS queues that fail persistently, stopping continuous failed polling and log noise.

• Added support for L40S GPU engines to the Track processing pipeline.

aiWare Engine Updates

• Fixed an issue where the SigLIP 2 engine produced shifted timestamps, dropped tail frames, and failed outright when extracting video keyframes.

• Fixed an issue where the SigLIP 2 engine's vision model was missing, causing indexed image and video vectors to be random, plus related text-processing padding and casing defects.

• Fixed a task failure in the Thumbnail Generator engine.

• Fixed an issue where tasks using the Amazon Text Recognition engine could become stuck in a running state indefinitely.

• Fixed an issue where the Amazon Rekognition Face Detection engine intermittently failed tasks due to upstream service throttling.

• Fixed an issue where the Exif Extract engine failed to parse metadata from certain valid media files, and where it returned internal errors on a large share of otherwise valid images.

• Fixed an issue causing the Speechmatics Transcription (v6) English engine to fail nearly all tasks due to an expired license token.

• Fixed an issue where the face detection engine incorrectly retried image chunks smaller than the minimum supported size instead of failing immediately.

• Fixed a data race in the shared text-processing library that could cause incorrect language detection under concurrent load across translate, sentiment, and related engines.

• Fixed AWS Rekognition and AWS Transcription throttling responses being misclassified as internal errors instead of rate-limit errors, and deployed the corrected error classification to US-1 production.

• Updated the AWS Transcription engine to use the per-cluster default storage bucket and fixed a broken container build.

• Updated the DMH distribute engine to fetch and decrypt the Ayrshare profile key from core at task pickup instead of receiving it in the task payload, improving credential handling.

• Added build provenance information (build date, commit, branch) to engine task logs and failure messages to speed up triage of stale or misbehaving builds.

• Deployed the Ayrshare distribution engine to US-1 production.