Version 2.4.7#
A new release of Sherpa AI Server is available: Version 2.4.7.
1. Image recognition in PDF files#
A Settings button has been added to the Documents screen. It opens the PDF Recognition dialog, where the image processing mode can be selected:
- built-in Tesseract OCR;
- a saved OCR or multimodal model;
- image recognition disabled.
The PDF text layer is always used, while scans and meaningful images are additionally recognized. The setting is applied during the next document indexing or reindexing. Tesseract OCR is used by default for new accounts.
On the Models screen, the Create Model and Edit Model dialog now includes a Model Type field with the values Text, Multimodal, and OCR.
A method for full file recognition and indexing has been added for integrations:
POST /api/v1/folders/{folder_guid}/files/{file_guid}/recognize
Administrators can also configure Tesseract OCR languages, page processing timeout, and maximum image size through environment variables.
An issue has been fixed where a successfully processed PDF could receive the Error status.
This helps index scanned documents and PDFs with images using built-in recognition or a selected model.
2. Improved DOCX indexing#
Processing of DOCX documents in different standard variants, Strict and Transitional, has been fixed. Sherpa AI Server now correctly extracts the main text, headers and footers, and footnotes.
For empty and corrupted documents, a controlled error is returned instead of an incorrect indexing result. No new interface elements were added for this change.
This improves compatibility with documents created in different versions of office applications.
3. Default mode for new chats#
On the Chat screen, the Default Chat Mode field has been added to the Default Chat Settings dialog.
The account administrator can now define which mode will be used when new chats are created. The Agent mode is selected by default.
This simplifies workspace setup and removes the need for users to choose the primary mode manually each time.
4. Personal chats open by default#
When signed in as a user, the Personal tab is selected by default on the Chat screen.
A new chat is created according to the current tab: personal on the Personal tab, shared on the Shared tab.
This reduces the risk of accidentally creating a shared chat and makes personal conversations more convenient to work with.
5. Attachments of any type in chat#
Files of any type can now be attached on the Chat screen. Sherpa AI Server automatically indexes only supported formats, but other attachments are no longer excluded from the message.
This helps send files together with a message even when their format is not supported by standard document indexing.