AI runtime release notes

AI runtime is the software layer that processes prompts and returns results. AI runtime includes the LLM, prompt templates, and processing pipelines that power your app’s intelligence features.

Changes to AI runtime can impact app performance. For each app version you create, you choose how AI runtime updates occur: either automatically when new versions are released, or when the current version expires, which allows time for testing and manual update.

For details about AI runtime, see AI runtime versions. For details about release and expiration dates, see the Release and support timeline.

2.0.1

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.

2.0.0

Automation projects in organizations with appropriate contract terms in place now offer agent mode. Agent mode processes documents with next-generation models for enhanced extraction accuracy and simplified configuration. Existing projects can update from legacy mode to agent mode.

Projects in agent mode can use a model to autogenerate classes and up to 20 fields based on uploaded documents.

For the AI runtime 2.0 release, automatic updates are temporarily disabled: apps on 1.x update to 2.x only when 1.x support expires. Major runtime updates typically require field and prompt adjustments, so create a new app version rather than updating existing app versions in place. AI runtime 1.x expiration is calculated based on agent mode availability in your specific tenant environment.

1.0.13

Table detection is improved, especially for Excel files and multipage tables in PDFs.

1.0.12

Automation projects and apps can now factor in more input when processing documents, improving accuracy across all model-based processes.
Table detection and extraction accuracy is improved for tables spanning multiple pages.
Processing for Excel files is optimized for improved extraction accuracy.

1.0.11

Cleaning functions applied to table extraction fields now process large tables as a single unit, enabling global operations like sorting across all rows.

1.0.10

In documents with many tables, table detection and extraction accuracy is improved by capturing additional metadata to better differentiate similar tables.

1.0.9

For text extraction fields, if applicable results are found in several locations in a document, the result with the highest field confidence score is now used. Previously, the first result was used regardless of confidence.

1.0.8

In projects that don’t use file splitting, you can reference file extensions to help classify documents. For example, the description for an Images class might be Files with image file extensions, like JPEG, PNG, and TIF.
Back-end logic that handles table extraction is enhanced to support more complex queries and increase response reliability through automatic retry methods.

1.0.7

The default models used for classification are upgraded.

1.0.6

There are no changes in this version.

The space normalization feature originally introduced in this version was reverted due to compatibility issues with table formatting in document reasoning fields.

1.0.5

Improvements to prompt templates add support for multi-row headers in spreadsheets.

1.0.4

The standard and advanced models used in automation projects are upgraded. The upgraded models showed modest but consistent accuracy improvements in internal benchmarking.

1.0.3

Processing for derived fields is improved to better handle references to table extraction fields.

1.0.2

Improvements to prompt templates add support for Excel source files that contain multiple small- to medium-sized tables on one sheet (totaling 200 rows and 30 columns), or one large table up to 80 columns.

1.0.1

Existing apps are locked to their current AI runtime, either 1.0.0 or, if the app is created with AI Hub version 24.48, 1.0.1.
Improvements to prompt templates provide better results for form-like documents, especially those that include checkboxes.

1.0.0

Initial AI runtime version that bundles all AI capabilities and updates delivered through AI Hub version 24.46. Apps created with AI Hub version 24.46 or earlier report using AI runtime 1.0.0, though their actual AI capabilities vary based on the specific AI Hub version used to create the app.

AI runtime is the software layer that processes prompts and returns results. AI runtime includes the LLM, prompt templates, and processing pipelines that power your app’s intelligence features.

For details about AI runtime, see AI runtime versions. For details about release and expiration dates, see the Release and support timeline.

2.0.1

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.

2.0.0

Automation projects in organizations with appropriate contract terms in place now offer agent mode. Agent mode processes documents with next-generation models for enhanced extraction accuracy and simplified configuration. Existing projects can update from legacy mode to agent mode.

Projects in agent mode can use a model to autogenerate classes and up to 20 fields based on uploaded documents.

For the AI runtime 2.0 release, automatic updates are temporarily disabled: apps on 1.x update to 2.x only when 1.x support expires. Major runtime updates typically require field and prompt adjustments, so create a new app version rather than updating existing app versions in place. AI runtime 1.x expiration is calculated based on agent mode availability in your specific tenant environment.

1.0.13

Table detection is improved, especially for Excel files and multipage tables in PDFs.

1.0.12

Automation projects and apps can now factor in more input when processing documents, improving accuracy across all model-based processes.
Table detection and extraction accuracy is improved for tables spanning multiple pages.
Processing for Excel files is optimized for improved extraction accuracy.

1.0.11

Cleaning functions applied to table extraction fields now process large tables as a single unit, enabling global operations like sorting across all rows.

1.0.10

In documents with many tables, table detection and extraction accuracy is improved by capturing additional metadata to better differentiate similar tables.

1.0.9

For text extraction fields, if applicable results are found in several locations in a document, the result with the highest field confidence score is now used. Previously, the first result was used regardless of confidence.

1.0.8

In projects that don’t use file splitting, you can reference file extensions to help classify documents. For example, the description for an Images class might be Files with image file extensions, like JPEG, PNG, and TIF.
Back-end logic that handles table extraction is enhanced to support more complex queries and increase response reliability through automatic retry methods.

1.0.7

The default models used for classification are upgraded.

1.0.6

There are no changes in this version.

The space normalization feature originally introduced in this version was reverted due to compatibility issues with table formatting in document reasoning fields.

1.0.5

Improvements to prompt templates add support for multi-row headers in spreadsheets.

1.0.4

The standard and advanced models used in automation projects are upgraded. The upgraded models showed modest but consistent accuracy improvements in internal benchmarking.

1.0.3

Processing for derived fields is improved to better handle references to table extraction fields.

1.0.2

Improvements to prompt templates add support for Excel source files that contain multiple small- to medium-sized tables on one sheet (totaling 200 rows and 30 columns), or one large table up to 80 columns.

1.0.1

Existing apps are locked to their current AI runtime, either 1.0.0 or, if the app is created with AI Hub version 24.48, 1.0.1.
Improvements to prompt templates provide better results for form-like documents, especially those that include checkboxes.

1.0.0

Initial AI runtime version that bundles all AI capabilities and updates delivered through AI Hub version 24.46. Apps created with AI Hub version 24.46 or earlier report using AI runtime 1.0.0, though their actual AI capabilities vary based on the specific AI Hub version used to create the app.