Dalet Flex 2026.10.0 Release Notes – Support Hub

Dalet is excited to present the release of Dalet Flex 2026.10.0.

This release of Dalet Flex provides the new features, quality and security enhancements, bug fixes, and other changes listed below.

Technical Information and Versions

Required Flex Xtend Versions

See this article Dalet Flex Xtend Versions to find the correct versions for working with Flex Xtend with Dalet Flex 2026.10.0.

Dalet Flex

New Features

Open Wizards in a new tab

To allow you to continue working after launching a wizard, from this version of Dalet Flex, you can configure a wizard to open in a new tab rather than in the tab that you were working in. You can also set the wizard to open any site as per the external URL in the configuration.

In the Wizard configuration screen for creating an External Wizard, a new single- field option Launch in has been added

Select Iframe within modal dialog to open the wizard in the same tab as previously. This is the default behavior.

Select New foreground browser tab to open the wizard in a new tab.

NOTE: Wizards are launched in the same way as before but the external url does not necessarily have to be a wizard. It can be any site as per ‘External Url’ defined in Core.

Star Review sessions

From this version of Dalet Flex, you can star a review session in FlexREVIEW and it will be added to the list of Starred items in Flex MAM.

Select the starred review in FlexMAM. Click Start in The Review panel displayed in FleMAM to start the Review session.

Click the 3-dot menu on items in the Starred list to reorder them.

Extract Keyframes from growing/live assets

From this version of Dalet Flex, Extract Keyframe plugin supports extracting keyframes from growing/live files. The Extract Keyframe plugin version has been updated to 3.6 to support this functionality. Keyframes are registered incrementally in batches while the job is still running. By default only complete batches of 20 keyframes are registered. Any unregistered extracted keyframes are registered once the job finishes,

NOTES

FIXED_COUNT is not supported for growing sources because the final duration/content of the source is not yet known. If the Fixed Number is set for the action, when the action is used on a growing source,

The job is rejected by FSP with this error Frame selection mode FIXED_COUNT is not supported for a growing source.

A live or dynamic DASH manifest cannot be processed because its duration is unknown. The FSP job fails with this Error Unable to extract frames, duration is unknown!

Display highlights on search results in the display template

From Dalet Flex 2026.10.0, the search results list highlights the searched term in the columns of the current template, in Table and Tile View.

By default, Highlight Search is disabled when you login to FlexMAM. To enable/disable highlighting, enable/disable the Highlight search option in the Settings pane. When Highlight Search is enabled, it persists throughout the session until you logout.

A highlight is displayed for search results that match string query values/numeric text added in the header search bar or in a filter chip. A highlight is displayed for a match anywhere within the word i.e. beginning, middle, inside a written word/sentence, end.

Highlights are only displayed in the Search Results list, they are not displayed in the Summary panel

NOTE: When Semantic Search is enabled, the Highlight search option is disabled.

Support Text LLM Chat

This version of Dalet Flex introduces support for general-purpose Text LLM Chat in dalet-ai-service. This enables creating workflows comprising prompt-driven text generation tasks, such as summarization, translation, rewriting, classification, and decision-making without external cloud dependencies.

Synchronous text chat via POST /text/chat for immediate, single-turn text generation.
b via POST /jobs (using jobInputType: "LLMJob"), allowing long-running or batch tasks to be queued and polled via GET /jobs/{jobId}.
Multi-tier model deployment using Google's Gemma 4 family across three selectable tiers: Fast (gemma-4-E4B-it), b(gemma-4-26B-A4B-it), and Best (gemma-4-31B-it), with optional reasoning mode (thinking) on supported tiers.

API for Synchronous Text Chat

The new endpoint POST /text/chat has been added.
Accepts input (text to operate on), prompt (instruction/task), optional tier (fast, balanced, best; defaults to balanced), and optional thinking boolean flag (defaults to false).
Returns HTTP 200 with model and generated output.

Example

curl --location '<dalet-ai-service base url>/text/chat' \
--header 'Content-Type: application/json' \
--data '{
    "tier": "balanced",
    "thinking": false,
    "input": "Dalet Flex is a media asset management and workflow orchestration platform that helps content creators and broadcasters manage complex production workflows.",
    "prompt": "Summarize the key value proposition in one sentence."
}' | jq .
API for Asynchronous LLM Job

Sample Response


{
 "model": "google/gemma-4-26B-A4B-it",
 "output": "Dalet Flex streamlines and manages complex production workflows for broadcasters and content creators as a unified media asset management platform."
}

The job type LLMJob (with jobInputType: "LLMJob") is now supported under POST /jobs.

Job execution runs in the background queue worker and results can be polled via GET /jobs/{jobId}.

Upon completion, the result is available in jobOutput under the same schema (model and output).

Example

curl -X POST --location '<dalet-ai-service base url>/jobs' \
--header 'Content-Type: application/json' \
--data '{
  "externalId": "Chat-Job-12345",
  "job": {
    "jobType": "AIJob",
    "jobProfile": "LLMJob",
    "jobInput": {
      "jobInputType": "LLMJob",
      "tier": "balanced",
      "thinking": false,
      "input": "Long transcript or document text goes here...",
      "prompt": "Extract key action items and decision points as bullet points."
    }
  }
}'
Job Polling (GET /jobs/{jobId})
{
  "id": "12345",
  "status": "COMPLETED",
  "dateCreated": "2026-09-28T12:00:00Z",
  "dateModified": "2026-09-28T12:00:15Z",
  "jobOutput": {
    "jobOutputType": "LLMJobOutput",
    "model": "google/gemma-4-26B-A4B-it",
    "output": "- Action item 1...\n- Decision point 1..."
  },
  "error": ""
}

General Details (applicable for both APIs)

Details on the supported model tiers:

Fast (google/gemma-4-E4B-it): 4.5B effective parameters. Best suited for high throughput, short outputs, and simple rewrites. Operates on CPU or smaller GPU footprints. Default max input tokens: 120,000.

Balanced (google/gemma-4-26B-A4B-it): Mixture of Experts (3.8B active / 25.2B total). Default tier for summarization, translation, and structured output. Requires a single mid-range GPU. Supports thinking mode. Default max input tokens: 250,000.

Best (google/gemma-4-31B-it): Dense 31B parameters with 256K context support. Best suited for deep reasoning, complex instructions, and large context synthesis. Requires a single high-end GPU. Supports thinking mode. Default max input tokens: 250,000.

Consul Configuration Keys

Per-tier settings can be configured at flex/dalet-ai-service/text/tiers/<tier>/config (where <tier> is fast, balanced, or best) with JSON fields:

model: HuggingFace model path (defaults: google/gemma-4-E4B-it, google/gemma-4-26B-A4B-it, google/gemma-4-31B-it)
maxInputTokens: Maximum accepted input tokens including prompt (default: 120000 for fast, 250000 for balanced/best)
maxNewTokens: Token generation budget (default: 10000)

Specific tiers can be enabled or disabled per node using flex/dalet-ai-service/text/tiers/<tier>/enabled (true/false).

Model Access / Hugging Face

Downloading Gemma 4 models requires HuggingFace access. Ensure a valid token with model access permissions is configured under flex/dalet-ai-service/huggingface/token (or HF_TOKEN environment va

Known issues

When running on CPU or low-resource nodes, initial synchronous requests to /text/chat will experience latency as model weights are loaded into memory. In K8s environment, for synchronous endpoint, you might have to increase the ingress timeout to allow the responses to be formed.

Enabling thinking mode on the fast tier is not supported and will result in an HTTP 400 error ("Thinking mode is not available on the 'fast' tier.").

Caveats

Hardware & VRAM Footprint

The balanced (gemma-4-26B-A4B-it) and best (gemma-4-31B-it) models require dedicated GPU hardware (minimum 16–24GB VRAM for balanced, and 32–48GB+ VRAM for best) in addition to base host memory.

Nodes without sufficient GPU resources should set flex/dalet-ai-service/text/tiers/best/enabled to false to avoid out-of-memory errors.

Out of Scope for v1

Multi-turn conversational state is not maintained server-side; chaining calls should be handled workflow-side.

Streaming responses, multi-modal input (images/audio), and tool/function calling are not part of this release.

Ingest Portal

New Features

Display a thumbnail on the Job Card in Ingest Portal Timeline

From this version of Dalet Flex, a thumbnail image is displayed on the Job Card for Recording/Completed jobs. The image is displayed on Job Cards that are 90px or larger. For smaller Job Cards, the image is hidden.

NOTE: The image is only displayed if thumbnails have been manually uploaded to the asset. Currently they are not generated dynamically.

Changelog

2026.10.0

Enhancements DaletFlex

To improve performance when working with WAV files in Adobe Premiere, Flex xTend creates DLT files when importing WAV files to Premiere.
FSP outputs segmented DASH/HLS generation for non-growing MXF.
flex-migrator: Bulk Migration for Flex Metadata 5
flex-migrator: Migrate all metadata definitions when no metadata definition is provided as a parameter in the migration.
Remove the domainname system property and derive account host URLs from Consul KVs.
Support the creation and use of DASH/HLS proxies that have subtitles (WEBVTT or TTML).
Update metadata instance documents in MongoDB to store userID and modifiedDate whenever an instance is saved.
Ability to cancel the dalet-ai-service jobs.
Growing File Support: Add the common polling and progress infrastructure to support JEF copy of growing files.
JEF Copy plugin: Update file size on polling ffpto support seeing the copied asset and its current size without waiting for the source file to finish rendering.
Add new channelMapping media profile API support . Legacy audio streamMap+channelMix are still supported.
Conform plugin: Fix the Transcode/Conform plugin's Extract Subtitles handling to support 0-to-many caption/subtitle caption extraction returned by DaletCUT.
In dalet-tenant-provisioning-service, it should not be mandatory to provide S3 access/secret keys as Consul KVs when an IAM instance role has been bound to the instance where it runs.
To avoid index locking, add a KV to skip embedding generation for Keyframes.

Flex/shared/featureToggles/search/ai/includeKFForEmbeddings = false – no image embeddings added

Flex/shared/featureToggles/search/ai/includeKFForEmbeddings = true – image embeddings added

Default KV value when not set = false. Invalid value = false. Image assets index their main image embedding.

When KV=false, reindexing an asset that has keyframe embeddings removes them, but text embeddings are unaffected.

Define FMP header permissions assigned to third party users created in Secure Share workflows.
Flex Xtend: Improve performance by reducing the authentication token generation rate.
Improve error logs for AZBS provider.
Support UTF-8 character rendering for conform jobs using Chinese text.
Provide Transcode audio mapping information.

Fixed Issues DaletFlex

Fixed an issue where alpha layer transparency is not preserved properly for marker icons.
Fixed an issue where markers icon do not support alpha-transparency channel in gif/png.
[Review Session]: Fixed an issue where Expression support for Priority variable in image and video review session actions was not working.
[Flex Xtend]Fixed an issue where selecting another project before export provokes: EvalScript error
[Flex Xtend] Fixed an issue where render send to Remote AME fails with error: Failed - unable to get local issuer certificate.
Xtend proxy mode: Fixed an issue in where assets become unlinked after saving and re-opening projects, especially when re-launching after several hours. Multiple error pop-us are displayed.
Fixed an issue with Conform plugin captions extraction by adding support 0 to many.
[Flex Xtend] Fixed an issue where XTEND Panel slows down when local storage contains many files.
Fixed an issue where flex-authentication-service failed migration.
Flex - Conform with title fails with preset conform-high-res. Fixed an issue where performing conform on a Flex DaletCUT failed when using the conform-high-res preset even though it succeeds when using other profiles.
Unable to upload image using 'Upload Thumbnail' wizard.
Fixed an issue where Import Project & Import Edit is slow to complete (up to 2 minutes).
Fixed an issue where API managed query causes FQL/OpenSearch pre-search and this causes paging failures when the offset reaches 10,000.
[flex-enterprise]: Fixed an issue where it was not possible to create Object fields of type UDO in metadata designer due to a missing id in object type uri. The UserDefinedObjectType API response now contains id, name, hrefdetails.
Fixed an issue where the tooltip is not displayed on mouse hover for long values in markers listing table on Summary panel.
Fixed an issue where the error The physical file cannot be accessed message is displayed when deleting an asset and moving it to a GCS storage resource config path that has space in its folder name even though the file has been moved to the deleted storage path.
Fixed an issue where Web Transfer import is not completing successfully for a GCS storage resource in the inbox configuration with space in front and at the end of folder name + file name.
Support starred item cleanup does not work for closed/expired review session.
FlexADMIN: Fixed an issue where Edit and Cancel icons in the Ingest tab of Application Management are too large
Growing File: Fixed an issue with the display of waveforms of growing assets.
Fixed an issue where vertical splitter resets to its original position when playing a video after navigating from the Audio Manage page to the Summary page and back again to the Audio Manage page.
Fixed an issue in Tile view where icons and text overlap.
[Manage Page] Fixed an issue where a grey bar is displayed on player waveform area.
[Ingest Portal] Add a space between Main and Location in the MainLocation label in the Add Recording dialog.
Fixed an issue where Third-party user access is denied for users with an email with underscores because the system does not process them correctly.
[Global Search] Fixed an issue where Global Search performed from Manage page returns previous searched result.
[Ingest Portal] Crash recording/scheduling job options should be disabled for FIFO configured target. FIFO logo is always displayed on a FIFO channel.
Fixed an issue where an action assigned to a group is not available in the Action Object Variable dropdown in Core and FlexMAM for group members.
Fixed an issue with duplicate Subjects handling and missing object types in Type dropdown in Subject listing page.
Fixed an issue where an incorrect error message is displayed when removing subjects from a deleted Role.
Fixed an issue where there is no preview whendragging and dropping an EDL created by Cut API.
[Xtend Flex] Job monitor: Fixed an issue where remote AME machine seems to no be connected and clear jobs is not possible.
Support expression for priority variable in image and video review session actions.
[Flex Xtend] Fixed an issue where the lock on an Adobe prproj does not get released when it is released on a different client machine.
Master Account Objects not available as subjects for Subject Scoped Role in Tenant Account.
Fixed an issue where Conform Plugin is failing in FILE Protocol from FSP.
[FQL] Fixed an issue where FQL is generating duplicate clauses.
Ingest Portal BFF: Fixed an issue where site cache shares sites and permissions across users.
Add security policy exception for flex-panels to allow image upload for thumbnails.
Fixed an issue where whitelisted extensions cannot be uploaded because the whitelisted extensions set in Consul are not being respected.
Fixed an issue where markers are not displayed on timeline.
[Xtend Flex] Fixed an issue where importing asset with an apostrophe failed with an EvalScript error.
Fixed an issue where Tesla P40 could not run transcription.
Fixed an issue so icon transparency is preserved for marker icons.
Fix broken button styles.
Fixed an issue where the blank/cross icon is broken in Metadata Definition sub-views.

List of Services upgraded between 2026.9.0 and 2026.10.0

For the new 2026.10.0 release, most services have changed.