Versions v41.1.x are available to SaaS customers only.
41.1.1 (8 May 2025)
Updates
This version includes a number of updates that optimize our internal testing and deployment processes.
41.1.0 (7 May 2025)
Training Data Management (TDM)
New
Search by Text Segment — We’ve implemented the Search by Text Segment functionality in Training Data Management for Field Identification models.
Previously, this capability was only available during Field Identification Supervision tasks. Now, users can search for fields by specific text segments directly within the Training Data Management interface. This update makes it easier to locate, review, and annotate data during the model training process.
To learn more, see Training Data Management.
Updated
TDM for Classification Interface enhancements — We’ve improved the TDM for Classification interface as part of our ongoing efforts to streamline the user experience:
The Training Status column has been replaced with Date Deployed in the Classification Models table, providing clearer insight into the latest deployed version.
The Last Training value has been removed from the Summary card in the Overview tab of TDM for Classification.
Document Drift Management (Layout Triage)
Updated
Bounding Box Overlay — We've introduced a bounding box overlay feature in Document Drift Management (Layout Triage), making it easier to verify that the document is a variation on an existing layout.
Click the Toggle Field Overlay button on the Edit Metadata step in the Layout Triage experience to enable the Bounding Box Overlay feature.
When enabled, this feature displays:
Blue bounding boxes representing field locations.
Field names as tooltips when hovering over a box.
Similar Layout Variation Detection — We’ve improved support for Similar Layout Variation Detection when the Validate Classification Using LayoutID setting is enabled in the flow.
Previously, when this setting was enabled, Similar Layout Variation Detection only considered matches where the layout identifier exactly matched the one specified in the Classification settings. To learn more, see Layout Identifiers and Structured Classification and Layout Identifiers.
If the layout identifier didn’t match—even if the document was visually identical to an existing layout—no suggestions were provided. With this update, Similar Layout Variation Detection now works even when Layout ID validation is enabled.
Note that this update does not modify the existing Layout ID validation logic in Classification. It only extends support for Similar Layout Variation Detection in Document Drift Management (Layout Triage).
Page preview pane added to the Page Groups tab — We've added a Page preview pane to the Page Groups tab in Document Drift Management (Layout Triage).
You can now preview pages directly from the Pages column in the Page Groups list, similar to the preview functionality already available in the Unmatched Submissions table. This update allows you to review pages directly in the tab.
Fixed
Messaging for training jobs in SaaS deployments — We've fixed an issue that caused the following message to appear on the model details page when the Run training button was clicked: "Trainer task created, but a model will not be trained because there is no trainer that matches this version." This issue occurred even when there was a trainer available.
Flows
New
Zooming in and out in Flow Studio canvas — You can now zoom in and out to adjust your view of a flow in Flow Studio. With this enhancement, you can switch between focusing on a specific part of the flow and viewing the flow in its entirety.
To zoom in or out, you can click the buttons in the upper-left corner of the canvas, or you can use the keyboard shortcuts Command + + to zoom in, Command + - to zoom out, and Command + 0 to reset zoom to 100% on Mac OS. On Windows, you can use Control + + to zoom in, Control + - to zoom out, and Control + 0 to reset zoom to 100%. You can also use touch gestures such as pinching with two fingers, or you can use your mouse's clickwheel while holding the Command key (on MacOS) and Control key (on Windows).
This feature makes working with the canvas more manageable when viewing large or complex flows.
Automatic data deletion for flow runs — Flow runs that are not initiated from a submission are now included for deletion in automatic data deletion. Automatic data deletion makes it easier to comply with security policies and manage storage space. This change expands Hyperscience's data-deletion capabilities, as policies for data deletion in previous versions applied only to submission-based flow runs.
Note that submission-based flow runs are not affected in the deletions scheduled as part of this feature.
For more information, see PII Data Deletion.
Tables
Fixed
Assigning child rows to new parent rows in nested tables — We've resolved an issue that prevented child rows from being assigned to new parent rows in nested tables. The issue occurred when the parent table spanned across two or more pages.
Extracting single-character values — We've fixed an issue that prevented the system from detecting and transcribing single-character values in table columns.
Custom Supervision
Updated
Document Chat flow enhancements — To improve the user experience for implementing Document Chat, we've made the following updates in this version:
"Custom SV with Chat" parameter — We've introduced a new parameter that makes it easier for a flow to take a Chat flow identifier as an input and, as a result, to include a Custom Supervision task configured with Document Chat in existing flows.
New
chatrolefor flows — Flow developers can now assignchatas arolefor Document Chat subflows, allowing these subflows to be distinguished on the Flows page and included as an option for the Chat Flow setting within Custom Supervision Blocks.
.png?sv=2022-11-02&spr=https&st=2025-05-11T11%3A24%3A56Z&se=2025-05-11T11%3A47%3A56Z&sr=c&sp=r&sig=VjxTQcLu0TopE%2BpwAR8iesFbWPH9qTaXXdorbh5GrmQ%3D)
For more details on Document Chat, see our Flows SDK documentation.
User Management
Fixed
Importing permission groups and authentication groups — We've fixed an issue that prevented users from importing both permission groups and authentication groups when no data for either type of group was already present in the instance (e.g., when migrating group data from one environment to another).
As part of this update, we've added a Create missing authentication groups setting to the Upload Existing tab of the Add Permission Group dialog box. When this setting is enabled, you can upload permission groups when there are no authentication groups present in the instance.
41.0.6 (24 Apr 2025)
VLM Blocks
Updated
Improved memory efficiency during Vision Language Model (VLM) initialization — We’ve reduced memory usage when initializing VLM Blocks. This update improves startup performance and stability by minimizing memory spikes when loading large model assets.
File Downloads
Updated
File-download behavior — We’ve updated the way files are delivered by the application. In certain cases, files may now be fetched directly from the object store using pre-signed URLs. While this change improves performance and scalability, download issues may occur in environments where object-store access is restricted at the network or browser level. To support these edge cases, you can revert to the previous file-delivery method by adding the following to your “.env” file:
REDIRECT_FILE_SERVING_TO_OBJECT_STORE=false
This setting ensures that files are downloaded and served directly by the application rather than accessed through the browser. For more information, please contact Hyperscience Support.
41.0.5 (14 Apr 2025)
Updates
This version includes a number of updates that optimize our internal testing and deployment processes.
41.0.4 (7 Apr 2025)
User Experience
Fixed
"Updated at" timestamps — We've fixed an issue that caused "Updated at" timestamps (e.g., in the upper-right corner of the Submissions and Flows pages) to be in the user's time zone instead of the time zone specified in the SYSTEM_TIME_ZONE ".env" variable.
Tables
Fixed
Processing submissions with nested tables — We've resolved an issue that caused submissions containing nested tables to halt in Machine Transcription. The issue occurred when all rows in a child table were deleted during Manual Identification.
41.0.3 (28 Mar 2025)
Reprocessing
Updated
"Reprocessing Enabled" setting — We've added a Reprocessing Enabled setting to the Reprocessing Block.
As part of this update, we've changed the logic for clicking Mark Layout Variation Incorrect during Supervision to the following:
If the Reprocessing Block is present and its Reprocessing Enabled setting is selected:
The layout will be removed from all pages of the current document, and a Manual Classification task will be created for it.
If the Reprocessing Block is not present, or if its Reprocessing Enabled setting is deselected:
Processing of the submission will end, and no data will be extracted from it.
41.0.2 (27 Mar 2025)
Updates
This version includes a number of updates that optimize our internal testing and deployment processes.
41.0.1 (18 Mar 2025)
Version 41.0.0 was not released and is not supported.
User Experience
Updated
Additional strings available for translation — We've enabled the translation of additional text in the application's user interface. This text is visible to users in the Data Keyers permission group (e.g., Supervision interfaces). For a complete list of text strings available for translation, download the en_US translation file after upgrading.
To learn more about translating the user interface, see Providing a Translated User Interface.
Layouts
New
Auto-splitting — Semi-structured documents now benefit from advanced layout-level grouping logic, enabling document-specific splitting rules. The Auto-splitting feature introduces enhanced document processing capabilities, allowing customers to split back-to-back documents of the same layout automatically. This update provides more flexibility and reduces manual intervention, improving efficiency for document-heavy workflows.
In v41, you can define specific rules for splitting documents within the Layout Editor. Only one rule can be applied per layout. The supported rules include:
Fixed Number of Pages — Automatically splits documents based on a set number of pages.
First Page Regex Match — Splits documents when a specific regular-expression (regex) pattern is detected on the first page.
Last Page Regex Match — Splits documents when a specific regex pattern is detected on the last page.
Page-to-Page Regex Matching — Compares regex patterns across consecutive pages to determine splitting logic.
If a splitting rule fails (e.g., regex not found or page count mismatch), documents can be automatically sent for manual review. This option is configurable at the layout level.
Auto-splitting is available for flows created in v40.2 and later.
For more information, see Auto-splitting.
Fixed
Marking of annotations as ineligible — We’ve resolved an issue where annotations were incorrectly marked as ineligible in Training Data Management when multiple live versions of the same layout existed.
Training Data Management
Updated
TDM for Classification user-interface enhancements — We’ve updated the user interface for managing Semi-structured Classification models to align with the user experience for Identification models. In particular, we’ve introduced the Model History table, which allows you to view and manage different versions of Classification models.
To learn more about TDM for Classification, see TDM for Classification Models.
Fixed
Previewing pages in Training Data Management (TDM) for Classification models — We've fixed an issue that caused delays in loading preview images of pages in TDM for Classification models.
Accuracy
New
Field- and column-level accuracy targets for identification in Semi-structured documents — In some submissions, identification accuracy may be more important for some fields or table columns than others (e.g., policy / account numbers, addresses). For these use cases, you can set individual accuracy targets for the identification of crucial fields and table columns. This update eliminates the need to set higher accuracy targets at the flow level, preventing your keyers from having to complete Identification Supervision tasks for lower-value fields and table columns.
You can set field- and column-level accuracy targets for identification in a flow's settings.
The accuracy achieved for each field or column you've set a specific target for can be found on the Accuracy page of the application (Reporting > Accuracy).
Note that setting accuracy targets at the field or column level has no effect on the number of QA tasks that the system creates.
To learn more about this feature, see Identification Settings.
Flows
New
Support for Python 3.12 — We now support the use of Python 3.12 in custom code, custom flows, and Python packages. In addition, we still support the use of Python 3.9 and 3.11 . This support ensures the continued security and reliability of your operations as you upgrade your code and flows to use Python 3.12.
The default flows included in Hyperscience have been updated to use Python 3.12.
While you can use any of our supported Python versions in Hyperscience v41, we recommend upgrading the entirety of your flows' code to use Python 3.12 as soon as possible.
As part of this update, a list of installed packages for Python 3.12 can be found on the Python Packages page in the application (Flows > Python Packages). The page also contains lists of installed packages for Python 3.9 and 3.11.
While Hyperscience supports the use of Python 3.9 in v41, note that Python 3.9 will reach end-of-life in October 2025 and will not be supported in v42.
More information about Python 3.12 support can be found in Developing Flows.
Updated
File Filter Block in Document Processing flow — The Document Processing flow included in v41 contains the File Filter Block, which allows you to specify the criteria for files that will be accepted or rejected during the submission-creation process. In previous versions of Hyperscience, this block could be used only in custom flows.
To learn more about the File Filter Block, see Document Processing Subflow Settings.
Fixed
Submission processing and size of database transactions — We've fixed a database-transaction issue that delayed the advancement of submissions through flows. As part of this update, we've introduced the HYPERFLOW_ENGINE_MAX_TASKS_PER_DB_TRANSACTION ".env" file variable, which allows you to specify the maximum number of workflow-engine tasks that are included in a single database transaction.
For more information, see Tasks Per Database Transaction.
LLM Blocks
Updated
LLM Install Block — With the LLM Install Block, you can use LLM Blocks in your own custom flows. When initiated, the LLM Install Block first checks for the presence of an LLM in the instance and installs one if one is not present. In previous versions, LLM Blocks could be used only in specialized flows provided by Hyperscience, and the model-installation process required a separate flow.
For more information on using the LLM Install Block, see Flow Blocks.
Custom Supervision
New
Document Chat — If your Knowledge Workers need specific, hard-to-find details to complete Custom Supervision tasks, you can add a Chat Component to a Custom Supervision Block. The Chat Component enables the Knowledge Worker to send questions or other prompts about the documents in the Custom Supervision Task to an LLM or other third-party service.
For example, a Knowledge Worker can ask, "Are there test results for [condition name]?" and have the LLM respond with relevant data from the document in complete sentences.
There are no restrictions on the prompts, or questions, that Knowledge Workers can send to the LLM. Because of the free-form nature of the prompts, we recommend making this feature available only to those Knowledge Workers who have training in the relevant field (e.g., nurses for medical decisions).
Note that this chat-like interface does not require an API connection to an LLM; if it is used with an LLM installed with the LLM Install Block, you can use the Chat Component in air-gapped instances that cannot access the internet.
Document Drift Management
New
Similar Layout Variation Detection — Similar Layout Variation Detection introduces an enhanced workflow for managing document variations by allowing you to easily identify and add new layouts as variations of existing ones. This update simplifies the process of handling diverse document formats and reduces the manual effort required to maintain layout variations. This feature is a key component of the broader Document Drift Management (Layout Triage) experience, aimed at improving document processing accuracy and efficiency.
Similar Layout Variation Detection is currently in beta. To enable this feature, or for more information, contact your Hyperscience representative.
Updated
User-interface enhancements — The Document Drift Management interface has been updated with improved navigation and visibility to support more effective user experience. To learn more, see Document Drift Management (Layout Triage).
Infrastructure
Updated
Supported Docker versions — Hyperscience v41 supports Docker 25.0.4 and later. It will not run on earlier versions. For more information, see Docker-Related Changes in v41.
Supported operating system versions — As part of v41, we've made the following changes to the versions of operating systems that we support:
Ubuntu — We've removed support for Ubuntu 19 and earlier.
Red Hat Enterprise Linux (RHEL) — We've removed support for RHEL 8.9 and earlier.
Supported database versions — In v41, we've made the following updates to the versions of databases that we support:
PostgreSQL — We've added support for PostgreSQL 15 and 16 and removed support for PostgreSQL 13.
Microsoft SQL Server (MSSQL) — We've added support for MSSQL 2022 and removed support for MSSQL 2016.
For information on supported container environments, operating systems, and databases, see Infrastructure Requirements.