Instabase Release Notes

Instabase Academy is now available, with persona-based video learning paths for developers and admins, as well as an AI Hub fundamentals course.

Automate, deploy, & review

Enterprise lite

Enterprise

• Resolved issue | App and deployment runs could intermittently fail with authentication errors due to stale Google authentication tokens. Runs now recover automatically from these failures.

API & SDK

• Resolved issue | When using the generate_post_process_pdf: true flag, generated PDFs had improperly oversized page dimensions, which could impact page-size-dependent workflows. PDF page sizes are now normalized.
• When using include_review_results=true for the Get run results endpoint, missing modifications[].message values are now returned as empty strings. Previously, missing values were returned as null. This change impacts messages in field, classification, and case-field edit histories (files[].documents[].fields[].edit_history[], files[].documents[].class_edit_history[], case_info.fields[].edit_history[]).
• Resolved issue | The above change to message values resolves an issue where the Python SDK raised an error and returned no results when run results included null messages.

Automate, deploy, & review

Enterprise lite

Enterprise

• Table fields in human review display letter column headers (A, B, C) for easier cell reference.
• Resolved issue | In human review, table cell validation error labels and highlights were off by one row when the first data row was used as column headers. Validation banners and jump-to-cell now match the displayed spreadsheet.
• Resolved issue | In human review, opening a table field for editing could fail after a large bulk cell edit.
• Preview | Resolved issue | For organizations with the Review workflow updates feature enabled, members with the tester or developer role had read-only access to review tasks. Members with these roles can now edit review tasks as expected.

API & SDK

• Single tenant | When using include_review_results=true for the Get run results endpoint, missing modifications[].message values are now returned as empty strings. Previously, missing values were returned as null. This change impacts messages in field, classification, and case-field edit histories (files[].documents[].fields[].edit_history[], files[].documents[].class_edit_history[], case_info.fields[].edit_history[]).
• Single tenant | Resolved issue | The above change to message values resolves an issue where the Python SDK raised an error and returned no results when run results included null messages.

Automate, deploy, & review

Enterprise lite

Enterprise

• The review interface has the following changes to the review task header:
  • The Submit and Next button is now either a Submit button (when you submit a single task or the final task in a series) or a Next button (when you submit one task in a series). Clicking Next both submits the task and moves you to the next in the series.
  • The back arrow button is replaced with an Exit button, which lets you exit a review session.
  • The Escalate button is moved to the center-left of the review task header.
  • When reviewing a series of tasks, a Task x of y indicator shows your progress.
• Preview | Human review has an updated workflow experience including role-based workflow views and timed task transitions. To use these features while in preview, turn on the Review workflow updates feature for your organization.
  • When submitting review tasks, a countdown displays rather than immediately advancing to the next available review task. Reviewers can choose to exit their session or pick up the next task. Previously, submitting a task automatically opened and began a review for the subsequent task.
  • Members with the reviewer or review manager role have access only to the Review tab in Workspaces. Members with the tester or developer role have read-only access to review tasks.
• Reviewers can bulk select and edit table cells in human review. Edits are applied to all selected cells.
• Human review has several performance improvements, including reductions in load times and lag, fewer page refreshes, and a smoother experience when reviewing long files and high-volume tasks.
• On the Review tab, clicking Refresh tasks also updates the overview counts for Not started, In review, and Completed tasks.
• If the deployment configuration for an in-progress run has been updated since the run started, the run status dashboard now indicates this under the Status column.
• Resolved issue | Retrying a deployment run after the deployment configuration was updated could cause the run to fail. Runs now retain the deployment version with which they were initiated.
• Resolved issue | When retrying a deployment or app run, tags weren’t preserved. Retried runs now retain the original tags.
• Resolved issue | On the Review tab, opening a task from the tasks list didn’t update its status from Not started to In review. Tasks now transition to In review when opened.
• Resolved issue | A header row offset caused table row indexing in human review to differ from the app editor, making cell-level validations harder to review. Indexing is now consistent across both.
• Resolved issue | In some deployments with escalation queues configured, the escalation option didn’t appear in human review.

Enterprise

• Resolved issue | In legacy flow job review, Excel files didn’t display when native Excel processing was enabled. The document viewer rendered blank and image load errors appeared. Excel files now display in the spreadsheet viewer.

This release includes security enhancements.

Automate, deploy, & review

Enterprise lite

Enterprise

• Preview | Single tenant | Human review has an updated workflow experience including role-based workflow views and timed task transitions. To use these features while in preview, turn on the Review workflow updates feature for your organization.
  • When submitting review tasks, a countdown displays rather than immediately advancing to the next available review task. Reviewers can choose to exit their session or pick up the next task. Previously, submitting a task automatically opened and began a review for the subsequent task.
  • Members with the reviewer or review manager role have access only to the Review tab in Workspaces. Members with the tester or developer role have read-only access to review tasks.
• Webhook notifications now support using Azure Event Grid payload formats. Payloads are sent as a single-element array with metadata for the selected format, CloudSchema or Event Grid. Preview and test notifications reflect your selection.
• Secret key names can now contain hyphens. This change adds support for webhook notifications to use hyphenated secret keys as custom headers.
• Resolved issue | For documents uploaded in non-PDF formats, such as .docx and .tiff files, long table and list extraction fields sometimes returned empty results. This issue could cause cascading errors and run failures. The processing pipeline now correctly handles non-PDF files.
• Resolved issue | When verifying ground truth values in human review, saving a correction could change table or list field types from text to table, causing revalidation to fail on unrelated fields that previously passed validation. Corrections now preserve the originally extracted field type.
• Single tenant | Human review has several performance improvements, including reductions in load times and lag, fewer page refreshes, and a smoother experience when reviewing long files and high-volume tasks.
• Single tenant | On the Review tab, clicking Refresh tasks also updates the overview counts for Not started, In review, and Completed tasks.
• Single tenant | If the deployment configuration for an in-progress run has been updated since the run started, the run status dashboard now indicates this under the Status column.
• Single tenant | Resolved issue | Retrying a deployment run after the deployment configuration was updated could cause the run to fail. Runs now retain the deployment version with which they were initiated.
• Single tenant | Resolved issue | When retrying a deployment or app run, tags weren’t preserved. Retried runs now retain the original tags.
• Single tenant | Resolved issue | On the Review tab, opening a task from the tasks list didn’t update its status from Not started to In review. Tasks now transition to In review when opened.
• Single tenant | Resolved issue | A header row offset caused table row indexing in human review to differ from the app editor, making cell-level validations harder to review. Indexing is now consistent across both.
• Single tenant | Resolved issue | In some deployments with escalation queues configured, the escalation option didn’t appear in human review.

Enterprise

• Packet-processing projects now support using custom functions to dictate intra-class prioritization.
• Intra-class prioritization was previously configured in project settings. You can now configure it on the Cross-class > Intra class tab while building your project.
• In packet-processing projects, class-level derived and custom function fields can now access extractions from every document in a class, not only the document selected by intra-class prioritization. Each class field argument includes an all_document_values list with every document’s extraction for that field; the other properties on the argument reflect the prioritized document. For details and an example, see Using class-level functions alongside intra-class prioritization.
• Resolved issue | In legacy flow job review, Excel files didn’t display when native Excel processing was enabled. The document viewer rendered blank and image load errors appeared. Excel files now display in the spreadsheet viewer.

Automate, deploy, & review

Enterprise lite

Enterprise

• In agent mode, the new Spreadsheet preprocessing project setting improves extraction for large CSV and Excel files. A dedicated processing pipeline delivers better accuracy and completeness for files that previously failed or returned silently truncated results when input or output token limits were exceeded. Requires AI runtime 2.0.1 or later.
• The Long table extraction field setting—available for list extraction and table extraction fields—now supports CSV and Excel files. When enabled, the field is extracted independently instead of sharing an output budget with other tabular fields, which improves completeness for long tables. This setting is complementary to but independent from the Spreadsheet preprocessing project setting. When the project-level setting is enabled, all large files are processed through the dedicated pipeline. When it’s disabled, the pipeline is enabled automatically for files that contain a long table field, but only for those fields; other fields continue to use the standard tabular processing flow.
• For details about when to use these settings and spreadsheet processing limits, see Spreadsheet limitations.
• Error reporting is improved for errors arising from custom classification functions.
• Resolved issue | In human review of advanced app runs, cross-class fields could appear in a different order than in the defined schema. They now display in the correct order.
• Resolved issue | After reclassifying a document in human review, field re-extraction failed for table extraction fields in legacy mode apps. Re-extraction now completes successfully for these fields.
• Resolved issue | In human review, document viewer zoom shortcuts failed for certain keyboard layouts, including EU keyboards. Zoom in and out with Ctrl + mouse wheel and Ctrl + + + / Ctrl + + / Ctrl + - now work as expected.
• Resolved issue | Accuracy tests could fail after re-extracting fields in a ground truth dataset, when field names contained spaces. Accuracy tests now complete successfully in these cases.
• Performance improvements reduce project loading times and validation run times.
• Resolved issue | In human review, revalidating a field could cause validation functions to fail if they referenced custom or secret keys. During revalidation, these key values are now passed to the function.

Enterprise

• In human review, corrections to document-level fields can now be applied to related cross-class fields. When a document-level change affects cross-class values, reviewers are now prompted to review and apply the update to all cross-class values.
• Related to this change, when reviewers correct a class field that feeds a ranked cross-class field, the corrected value now appears in the ranked field’s dropdown options. Previously, manually corrected input values weren’t shown in the dropdown at all.
• Resolved issue | On the Cross-class tab, certain fields could remain stuck in a loading state. Fields now load as expected.

Administration

Enterprise lite

Enterprise

• A new Advanced developer workspace role grants access to the flow editor and the ability to create and publish advanced apps. Assign it to technically proficient builders, such as solutions engineers, who need flow editor access without workspace manager permissions. For detailed permissions, see Organization and workspace roles.
• The advanced developer role appears in the workspace role picker for all organizations, but it only grants additional capabilities compared to developer in single-tenant, enterprise-tier organizations with advanced view enabled.
• There have been the following changes to role-based permissions:
  • Members with the tester role or higher can run apps and view runs and run logs. Previously available to the developer role or higher.
  • Members with the developer role or higher can edit and delete apps created by other members in workspaces where they have developer access, based on the app’s linked source project. Previously available to admins only.
  • Members with the workspace manager role or higher can assign workspace default drives and designate production workspaces. Previously available to the admin role only.

This release includes security improvements.

Automate, deploy, & review

Enterprise lite

Enterprise

• In human review, cross-class ranked fields now display class tags showing which document class each value came from. Reviewers can click the tag to open the source document and field to verify or correct the value.
• Resolved issue | When the extraction model returned None, fields could display ERROR as the value instead. Affected fields now display None or Value(None).
• Resolved issue | In projects containing extraction and derived or custom function fields, extracted values could be assigned to the wrong field. These incorrect values could be propagated to downstream steps such as cleaning functions. This issue didn’t impact app or deployment runs.
• Resolved issue | In review, .xlsx files extracted from .msg files didn’t display correctly.
• Single tenant | Resolved issue | After reclassifying a document in human review, field re-extraction failed for table extraction fields in legacy mode apps. Re-extraction now completes successfully for these fields.
• Single tenant | Resolved issue | In human review, document viewer zoom shortcuts failed for certain keyboard layouts, including EU keyboards. Zoom in and out with Ctrl + mouse wheel and Ctrl ++ / Ctrl + - now work as expected.
• Single tenant | Resolved issue | Accuracy tests could fail after re-extracting fields in a ground truth dataset, when field names contained spaces. Accuracy tests now complete successfully in these cases.

Administration

• Resolved issues | For apps created after March 16, 2026, runs of legacy mode apps were being incorrectly billed as automation project usage.

This release includes stability and performance improvements.

Automate, deploy, & review

Enterprise lite

Enterprise

• For table extraction and list extraction fields, a new Long table extraction field setting is available. This setting delivers improved extraction results for multipage tables, with a dedicated discovery and extraction pipeline. Table highlighting is also improved, with one bounding box per page on which the table is found, rather than one bounding box on the table’s starting page only. Available in agent mode.
• Deployment integration functions now support shared functions. Add a function to the function library, then import it to your downstream integration custom function.
• A Tags column in the review dashboard lets reviewers filter jobs by tags. Tags can be added to runs initiated via API.
• Resolved issue | Split classification could fail when the model returned page numbers beyond the document length, leaving you with no results for that file. Out-of-range page numbers are ignored or adjusted so classification completes.
• Resolved issue | In human review, files with a # in the file name failed to load.
• Resolved issue | In human review, the document viewer zoomed in automatically when selecting a field. The zoom level is now fixed and can only be adjusted manually.
• Resolved issue | During agent mode app runs, logs were flooded with ERROR-level messages with no user-facing impact. Log entries for expected conditions are now logged at the appropriate level, making logs easier to use for troubleshooting.
• Resolved issue | For list of objects fields, the extraction model sometimes omitted a nested column on individual rows, which misaligned values in downstream custom functions that process rows by column order. Every row now includes all nested columns declared in your schema, in consistent order.
• Resolved issue | For agent mode projects previously configured in legacy mode, legacy mode file processing settings could override agent mode processing settings. Agent mode processing settings now take precedence.
• Resolved issue | In projects with both file splitting and advanced checkbox extraction enabled, the text of the entire document was being passed to the model, rather than the text of the split file. For fields with the same name across classes, the model could return values from the wrong class.
• Resolved issue | Users could encounter issues making edits in human review for apps without validations.
• Resolved issue | When exporting and importing projects, custom functions created from validation prompts were recreated instead of preserved.
• Resolved issue | In review, .xlsx files extracted from .msg files didn’t display correctly.

Enterprise

• Single-tenant, Enterprise-tier organizations can request access to the flow editor, a new way to build advanced apps. The flow editor introduces a full workflow automation layer to AI Hub, letting technical users build complex end-to-end workflows.

Administration

Enterprise lite

Enterprise

• AI Hub’s subscription offerings are updated:
  • A new Enterprise lite tier is designed for focused functional teams, high-growth startups, and mid-market organizations looking to rapidly deploy powerful document automation. It provides the full power of the AI Hub engine but is streamlined to remove complex orchestration features not required for single-team or API-driven deployments.
  • Deprecation | The Commercial subscription tier is deprecated and new sign-ups are discontinued. Existing organizations on the Commercial tier can connect with their support team directly for any questions.
  • Removal | The Community tier is fully removed.
• Resolved issue | Changes to custom keys in projects and deployments weren’t recorded in audit logs, so you couldn’t tell who added, edited, or removed a key.

Automate, deploy, & review

Enterprise lite

Enterprise

• Error reporting is improved for errors arising from custom classification functions.
• Resolved issue | In human review of advanced app runs, cross-class fields could appear in a different order than in the defined schema. They now display in the correct order.
• Resolved issue | After reclassifying a document in human review, field re-extraction failed for table extraction fields in legacy mode apps. Re-extraction now completes successfully for these fields.
• Resolved issue | In human review, document viewer zoom shortcuts failed for certain keyboard layouts, including EU keyboards. Zoom in and out with Ctrl + mouse wheel and Ctrl + + / Ctrl + - now work as expected.
• Resolved issue | Accuracy tests could fail after re-extracting fields in a ground truth dataset, when field names contained spaces. Accuracy tests now complete successfully in these cases.
• Performance improvements reduce project loading times and validation run times.
• Single tenant | Resolved issue | In human review, revalidating a field could cause validation functions to fail if they referenced custom or secret keys. During revalidation, these key values are now passed to the function.

Enterprise

• In human review, corrections to document-level fields can now be applied to related cross-class fields. When a document-level change affects cross-class values, reviewers are now prompted to review and apply the update to all cross-class values.
• Related to this change, when reviewers correct a class field that feeds a ranked cross-class field, the corrected value now appears in the ranked field’s dropdown options. Previously, manually corrected input values weren’t shown in the dropdown at all.
• Single tenant | Resolved issue | On the Cross-class tab, certain fields could remain stuck in a loading state. Fields now load as expected.

Administration

Enterprise

• Single tenant | Preview | Group mapping is now supported with OIDC-based single sign-on (SSO). Previously, mapping was supported only with SAML-based SSO.
• To use this feature while in preview, turn on the Enable OIDC group mapping feature for your organization.
• Single tenant | When configuring group mapping, you can now map multiple identity provider (IdP) groups to one AI Hub group. Previously, each AI Hub group could map to at most one IdP-defined group.

This release includes security improvements.

Automate, deploy, & review

Enterprise lite

Enterprise

• In human review, cross-class ranked fields now display class tags showing which document class each value came from. Reviewers can click the tag to open the source document and field to verify or correct the value.
• Resolved issue | When the extraction model returned None, fields could display ERROR as the value instead. Affected fields now display None or Value(None).
• Resolved issue | In projects containing extraction and derived or custom function fields, extracted values could be assigned to the wrong field. These incorrect values could be propagated to downstream steps such as cleaning functions. This issue didn’t impact app or deployment runs.
• Resolved issue | In review, .xlsx files extracted from .msg files didn’t display correctly.
• Single tenant | Resolved issue | After reclassifying a document in human review, field re-extraction failed for table extraction fields in legacy mode apps. Re-extraction now completes successfully for these fields.
• Single tenant | Resolved issue | In human review, document viewer zoom shortcuts failed for certain keyboard layouts, including EU keyboards. Zoom in and out with Ctrl + mouse wheel and Ctrl + + / Ctrl + - now work as expected.
• Single tenant | Resolved issue | Accuracy tests could fail after re-extracting fields in a ground truth dataset, when field names contained spaces. Accuracy tests now complete successfully in these cases.

Administration

• Resolved issues | For apps created after March 16, 2026, runs of legacy mode apps were being incorrectly billed as automation project usage.

This release includes stability and performance improvements.

Automate, deploy, & review

Enterprise lite

Enterprise

• For table extraction and list extraction fields, a new Long table extraction field setting is available. This setting delivers improved extraction results for multipage tables, with a dedicated discovery and extraction pipeline. Table highlighting is also improved, with one bounding box per page on which the table is found, rather than one bounding box on the table’s starting page only. Available in agent mode.
• Deployment integration functions now support shared functions. Add a function to the function library, then import it to your downstream integration custom function.
• A Tags column in the review dashboard lets reviewers filter jobs by tags. Tags can be added to runs initiated via API.
• Resolved issue | Split classification could fail when the model returned page numbers beyond the document length, leaving you with no results for that file. Out-of-range page numbers are ignored or adjusted so classification completes.
• Resolved issue | In human review, files with a # in the file name failed to load.
• Resolved issue | In human review, the document viewer zoomed in automatically when selecting a field. The zoom level is now fixed and can only be adjusted manually.
• Resolved issue | During agent mode app runs, logs were flooded with ERROR-level messages with no user-facing impact. Log entries for expected conditions are now logged at the appropriate level, making logs easier to use for troubleshooting.
• Resolved issue | For list of objects fields, the extraction model sometimes omitted a nested column on individual rows, which misaligned values in downstream custom functions that process rows by column order. Every row now includes all nested columns declared in your schema, in consistent order.
• Resolved issue | For agent mode projects previously configured in legacy mode, legacy mode file processing settings could override agent mode processing settings. Agent mode processing settings now take precedence.
• Resolved issue | In projects with both file splitting and advanced checkbox extraction enabled, the text of the entire document was being passed to the model, rather than the text of the split file. For fields with the same name across classes, the model could return values from the wrong class.
• Resolved issue | Users could encounter issues making edits in human review for apps without validations.
• Resolved issue | When exporting and importing projects, custom functions created from validation prompts were recreated instead of preserved.
• Single tenant | Resolved issue | In review, .xlsx files extracted from .msg files didn’t display correctly.

Enterprise

• Single tenant | Single-tenant, Enterprise-tier organizations can request access to the flow editor, a new way to build advanced apps. The flow editor introduces a full workflow automation layer to AI Hub, letting technical users build complex end-to-end workflows.
• Resolved issue | In human review for advanced apps, the interface didn’t clearly indicate which files required review or had errors.
• Single-tenant | Deployment integration functions now support the file client (ibfile) and LLM client (llm_client) through context['clients']. The integration function header is updated accordingly. To learn more about the ibfile and llm_client clients, see Calling LLMs from custom functions.

API & SDK

• You can add tags to your deployment and app runs, letting you categorize runs. Possible uses include tagging testing or development runs. Add tags as an array under settings, with the combined character count not exceeding 500.
• The List runs, Get run status, Run deployment, and Run app endpoints now return more run details, including batch_id, input_dir, app_id, deployment_id, and tags.

This release includes stability and performance improvements.

Automate, deploy, & review

Commercial

Enterprise

• In agent mode, two new file processing settings are available. Review the technical details and limitations before enabling in your project.
  • OCR-based provenance improves how extracted text is highlighted in a document, using OCR-based bounding boxes for more precise results than the default.
  • Large file extraction mode improves classification and extraction for large documents when inputs exceed typical model context limits. It’s most useful beyond 200 pages, and for documents over 1,000 pages where you might see missing fields or lower accuracy.
• Resolved issue | In agent mode, when classification split a multi-record PDF into separate PDFs, extraction could return no values for any field because the extraction prompt still referenced page numbers from the original document.
• Resolved issue | App runs could fail for some older app schemas, due to a capitalization mismatch between actual and expected values in the schema. The affected schema fields are now compared case-insensitively.
• Single tenant | Resolved issue | Split classification could fail when the model returned page numbers beyond the document length, leaving you with no results for that file. Out-of-range page numbers are ignored or adjusted so classification completes.
• Single tenant | Resolved issue | In human review, files with a # in the file name failed to load.
• Single tenant | Resolved issue | In human review, the document viewer zoomed in automatically when selecting a field. The zoom level is now fixed and can only be adjusted manually.
• Single tenant | Resolved issue | During agent mode app runs, logs were flooded with ERROR-level messages with no user-facing impact. Log entries for expected conditions are now logged at the appropriate level, making logs easier to use for troubleshooting.
• Single tenant | Resolved issue | For list of objects fields, the extraction model sometimes omitted a nested column on individual rows, which misaligned values in downstream custom functions that process rows by column order. Every row now includes all nested columns declared in your schema, in consistent order.
• Single tenant | Resolved issue | For agent mode projects previously configured in legacy mode, legacy mode file processing settings could override agent mode processing settings. Agent mode processing settings now take precedence.
• Single tenant | Resolved issue | Users could encounter issues making edits in human review for apps without validations.

Enterprise

• AI Hub’s subscription offerings are updated:
  • A new Enterprise lite tier is designed for focused functional teams, high-growth startups, and mid-market organizations looking to rapidly deploy powerful document automation. It provides the full power of the AI Hub engine but is streamlined to remove complex orchestration features not required for single-team or API-driven deployments.
  • Deprecation | The Commercial subscription tier is deprecated and new sign-ups are discontinued. Existing organizations on the Commercial tier can connect with their support team directly for any questions.
  • Removal | The Community tier is fully removed.
• Resolved issue | Changes to custom keys in projects and deployments weren’t recorded in audit logs, so you couldn’t tell who added, edited, or removed a key.

Automate, deploy, & review

Commercial

Enterprise

• In this release, how you access and navigate to reviews has changed.
  • The format for the URLs at which run-based reviews are accessed is changed.
    • Deployment runs: .../review/?deploymentId= → .../preview/
    • App runs: .../review/ → .../preview/
  • The Documents view on the Review tab is removed, in favor of a task-based review system that supports packet-processing apps. All existing document-based reviews have been migrated to tasks. The URL format for task-based reviews is .../review/task/.
• You can export and import automation projects as ZIP files to move work across workspaces, organizations, and environments, or to create backups. Exports include the project details, schema, validation rules, and custom functions, including shared functions from the function library. Sample files or previously uploaded files aren’t included.
• In agent mode, the new Advanced checkbox extraction file processing setting improves checkbox detection accuracy and can also improve results for table fields. Review the technical details before enabling in your project.
• Performance is improved when correcting tables in human review.
• The review dashboard has a new Date added column, letting reviewers sort and filter review tasks by date.
• You can create ground truth datasets from app runs that contained processing errors. The files that resulted in errors aren’t included in the dataset.
• Resolved issue | The number of fields per class was limited to 100, rather than the stated limit of 250. Adding 250 fields per class is now fully supported.
• Resolved issue | Hidden fields sometimes didn’t display correctly in human review.
• Resolved issue | Cleaning functions sometimes failed to rerun after reclassifying a document.
• Resolved issue | In human review, document sequence numbering was reversed. Reviewers now see the expected document 1 of 3 format, rather than document 3 of 1.
• Single tenant | Resolved issue | App runs could fail for some older app schemas, due to a capitalization mismatch between actual and expected values in the schema. The affected schema fields are now compared case-insensitively.
• Single tenant | Resolved issue | In agent mode, when classification split a multi-record PDF into separate PDFs, extraction could return no values for any field because the extraction prompt still referenced page numbers from the original document.

Enterprise

• The human review interface now fully supports packet-based review. Improvements to the review experience include:
  • Packet-processing app results can now exclusively be reviewed by packet. Previously, reviewing by file or by run was supported, which impacted visibility of cross-class fields. This change is also reflected in deployment configuration: for packet-processing apps, configuring Review by file or Review by run isn’t supported.
  • The review dashboard makes it clearer that you’re reviewing by packet, lets you filter review tasks by packet, and no longer requires a separate entrypoint for packet-based review.
  • When reviewing ranked cross-class fields, reviewers can select from a dropdown of extracted value options, or enter a custom value. The dropdown options show extracted values, not any manually corrected values. If reviewers have previously corrected the value of a class field, the dropdown options don’t reflect this change.

Enterprise

• Single-tenant | Resolved issue | In human review for advanced apps, the interface didn’t clearly indicate which files required review.

Automate, deploy, & review

Commercial

• Resolved issue | When copying projects across workspaces, custom functions weren’t copied correctly.
• In agent mode, the new Force image OCR file processing setting lets you treat all pages as images and process them with OCR. When enabled, text extraction from machine-readable PDFs is skipped, which can improve processing speed and reduce memory usage for certain documents.
• This release includes memory optimization improvements to reduce peak memory usage during file processing. Expect improved stability and reliability when processing large files in agent mode.

Enterprise

• In cross-class custom function fields, you can now set the return type to Text choices, limiting valid returned values to a specified set. If the function returns a value not in that list, a validation error is shown. Additionally, in human review, reviewers can search and select from the allowed list rather than typing values themselves.

This release includes security enhancements that disable local caching of potentially sensitive data.

Automate, deploy, & review

Commercial

• As previously announced, deployment integration functions now return results in the v2 format, which matches the structure returned by the Get run results API endpoint. To support this change, a migration helper is now available when editing existing integration functions. Use the helper to migrate your code and test changes before saving and permanently updating to the v2 format.

• Resolved issue | Users with tester permissions couldn’t view deployment configurations.

• The pop-out view for tables now displays wide, single-row tables more clearly.

• Single tenant | Resolved issue | When copying projects across workspaces, custom functions weren’t copied correctly.

Administration

Commercial

• When creating secrets, you can now use files as the secret’s value, such as uploading certificate files.

Enterprise

• When creating secrets, you can now use files as the secret’s value, such as uploading certificate files.

Automate, deploy, & review

Community

• For extraction fields, handling of unclear or unavailable values is improved. Previously, N/A was returned, which could conflict with valid extracted values. Now, No value found is returned by default. This change is available immediately in automation projects and existing community apps, and in commercial and enterprise apps as AI runtime 2.0.1.
• Resolved issue | Renaming projects sometimes failed.

Commercial

• File splitting now supports files up to 400 MB and 2,000 pages when the Disable OCR setting is enabled. After splitting, individual documents should be 200 pages or fewer for optimal extraction results. For details about requirements and constraints, see file limitations.
• Single tenant | You can call LLMs from custom functions in validation functions, cross-class validation functions, and cross-class custom function fields. The LLM client uses the environment’s configured LLM provider and the configured AI runtime model. Supported in agent mode only.
• Resolved issue | When attempting to convert generated validation code to a custom function, the confirmation modal would close before users could confirm the action.
• Resolved issue | Users with tester permissions couldn’t view a project’s settings.
• Resolved issue | Apps with associated ground truth datasets failed to delete, despite showing a confirmation message. Now when you delete an app, the app and all associated ground truth datasets are deleted properly.
• Resolved issue | In projects, custom validation functions showed as having failed when the validation result exceeded the character limit. Improved error messaging now clearly indicates this root cause. This limitation doesn’t affect apps or deployments.

Enterprise

• Run results sent via downstream integrations or retrieved through manual export now include packet-level information, including cross-class fields. Previously, packet-level information was available only when getting results via API.

Administration

• AI Hub now enforces stricter access controls for cross-domain integrations. By default, requests from external websites are blocked unless explicitly verified as trusted, preventing unauthorized websites from performing actions on behalf of logged-in users.

Action required: If your organization integrates AI Hub with other applications (for example, embedding AI Hub in a portal or accessing it from a different domain), you must connect with Instabase Support to provide a list of URLs of your trusted applications. Support can configure the appropriate settings to maintain connectivity. Otherwise, AI Hub now allows same-origin requests only.