What's New in DocCenter

4.5 release (9 October 2026)

DocCenter 4.5 introduces redaction models for permanently removing sensitive content from documents, and extends AI action reporting to verification models.

Redact sensitive content from documents

Redaction is now a top-level DocCenter model type. Redaction detects sensitive content—such as personally identifiable information (PII)—using keyword, regular expression, and semantic rules, then redacts it into a flattened PDF.

During reconciliation, you can review the AI-detected content to redact, and delete or draw boxes as needed. Once you complete the reconciliation tasks, redactions become permanent. Run redaction on its own or chain it before or after extraction and classification using triggers.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

Redaction addresses your compliance requirements and further enables you to process sensitive documents natively in DocCenter.

Monitor AI actions consumed by agents and verification models

DocCenter now reports AI action consumption for the features that use Appian agents, including verification models and AI-Suggested Improvements. With this change, you can track consumption for every DocCenter feature—extraction, classification, verification, and redaction—more consistently and transparently.

Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

Resolved issues

DocCenter 4.5 fixes a bug that caused large documents to fail extraction when using vision-enabled fields. These documents are now chunked correctly so extraction completes successfully.

What's new with processing version 4.5?

Beginning with DocCenter 4.4, you can choose to upgrade the processing version for individual models.

When upgrading to processing version 4.5, the following changes may affect your model's extraction and classification accuracy.

Classification changes

• Sonnet 4.6 now serves as the primary LLM, instead of Haiku 4.5.
• For new classification models, Vision Classification is enabled by default.

Extraction changes

• Advanced Layout has been removed as option for all new extraction models. It remains available for extraction models that use processing version 4.4 or earlier.
• Sonnet 4.6 now serves as the primary LLM, instead of Haiku 4.5.
• For new extraction models, Visual extraction is enabled for all fields by default.

Hotfixes for DocCenter 4.5

v4.5.1 (Oct 22, 2026)

• ADCS-1784 - High
  Fixed an issue where classifying .xlsx files failed with an "unsupported file type" error. Classification now completes for .xlsx files.

• ADCS-1776 - Medium
  Bounding-box rendering on image-heavy PDFs no longer inflates file size. Boxes are now drawn as vector overlays instead of re-encoding page images.

• ADCS-1841 - Low
  Fixed an error that occurred when sorting on the Instances column in Verification and Redaction model grids. Updated the column to sort on the universal instancesCount field.

• ADCS-1825 - Low
  Fixed an error that appeared when reconciling a classification instance on a model with more than 10 categories if the AI didn't select a category. The Category dropdown now shows a "Select a Category" placeholder instead of throwing an error.

Feedback