Skip to content

AIIDA Release Notes


  • Handling .heif/.heic/.avif file types in API and UI
  • Optimised image compression
  • New feature: Autobook toggle with Fortnox collections
  • Minor bugs and fixes


  • Improved extraction on scanned receipts
  • Improved learning on supplier level


  • Improvements for Total extraction
  • Improved support for highlighting extracted text in document



  • Improved extraction time
  • Improvements to online training


  • Improvements to extraction of total and vat
  • Interpretation speed improvements
  • UI updates regarding warnings on collapsable fields
  • Possibility to move items from UI (drag and drop) to collections and workflows via workspace list
  • Improvements to debit/credit setting for accountings


  • More accepted monetary values in interpretation and accounting fields
  • Minor bug fixes and improvements


  • New status page:
  • Dynamic status health on Aiida website
  • Minor bug fixes and improvements


  • Adding of more standard accounts
  • UI/UX updates regarding handling/labeling of items
  • Balancing of amounts now considered for an item's status
  • Simplified the creation of new collections with Fortnox
  • UX updates regarding interpretation and accounting fields
  • Updates to discrepancy calculations
  • Updates to the newly added collection classifier
  • Minor bug fixes and improvements


  • New custom Sharepoint export
  • Accountings with prefilled amounts, but no account number if no account was predicted
  • UX updates regarding interpretation and accounting fields
  • Fixes regarding item state
  • Possibility to move items between collections in UI (drag and drop) and via API added
  • Minor bug fixes and improvements


  • Fortnox OAuth2 integration
  • UX updates regarding interpretation and accounting fields
  • Minor bug fixes


  • Minor bug fixes


  • General speed improvements
  • Minor bug fixes


  • Account prediction improvements
  • New online trained accounting prediction
  • Offline interpretation algorithms improvements
  • New feature: Supplier id
  • Minor bug fixes


  • Improved account prediction
  • Interpreter algorithm improvements
  • Document classification improvements
  • Minor bug fixes


  • New offline trained interpretation algorithm (focus on VAT)
  • UX updates regarding status messages and accounting
  • Account prediction improvements
  • Zero VAT is not set on invoices with missing VAT
  • Overall responsiveness and speed improvements
  • Critical bug fixes regarding unsigned monetary field and account prediction
  • Minor bug fixes


  • New feature: Get and set predefined accounts on workflow level
  • New feature: Status messages for account prediction
  • Updated swagger documentation
  • Interpretation prediction algorithms improvements
  • Minor bug fixes


  • Better management of large documents
  • api-standard collection template now contain other
  • Offline trained interpretation algorithms improvements
  • Improved monetary field formatting
  • Improved extraction of texts in documents
  • Minor bug fixes


  • Items in process state will timeout after 30 minutes and set to error state
  • Upload items stabilization: Automatic retry of stuck and error items older than 10 mins and younger than 30 mins
  • New feature: Get collection endpoint
  • Negative monetary values are transformed to positive monetary values (absolute value)
  • Date and BIC field type improvements
  • Offline interpretation algorithm improvements
  • Minor bug fixes


  • Template type-a now uses "Credit invoice" workflow
  • Online/Offline interpretation improvements (better accuracy)
  • Field types improvements for Invoice number and IBAN
  • Minor bug fixes


  • Accept null setting when creating a collection
  • Apply templates will now merge with current collection and workflow
  • Workspace classification: Mail import for workspaces
  • BIC field part of reminder workflow
  • Skip duplicate status messages
  • Offline training improvements for classification, interpretation and accounting


  • Improved offline trained interpretation algorithm
  • New feature: Workspace classifier
  • New and improved OCR service


  • Finalize improvements
  • Added new field type BIC
  • Updated formatting and validation on IBAN and monetary fields
  • Updated collection templates
  • Minor bug fixes
  • Improve robustness of OCR service
  • Better logging for document with bad text layer
  • New interpretation algorithm specialized for scanned documents (receipts etc)
  • More robust interpretation
  • Improved online trained interpretation algorithm
  • New collection template
  • Updated API documenation
  • Minor bug fixes


  • New and better OCR service for non PDF documents and documents with bad text layer
  • Finalize item, easier and better management of training of items
  • New collection template ’type-a’
  • UX updates for warning messages
  • Minor bug fixes


  • Interpreter improvements regarding BG and PG (Less false positives)
  • Duplicate item warning. Status message will contain a warning message if invoice number and/or OCR number exist in another document (for same supplier)
  • Redo OCR correctly on PDF's with a specific type of bad text layer
  • API Documentation updated
  • Improved management of large documents


  • Offline trained interpretation algorithms improvements (less false positives)

  • Extraction of large documents speed optimization

  • Automatic addition of multiple marked VAT amounts


  • Minor bug fixes

  • Updated swagger documentation

  • Accounting improvements


  • New feature: Field status and warning messages

  • New feature: Dynamic document classifications

  • New feature: Finalize endpoint for easier integration

  • New collection template


  • User activation fix

  • Accounting improvements (improved accuracy)

  • Online trained interpreter improvements (improved accuracy)

  • Standard accounts for receipts

  • General speed optimizations

  • Field matcher improvements


  • Improved formatting rules for invoice numbers and monetary fields

  • Speed improvements regarding workflows


  • Improved offline trained interpreter (better accuracy)

  • Currency field improvements (less false positives)

  • General speed improvements


  • Improved offline interpreter algorithms (better accuracy and less false positives)

  • Improved document classifier

  • Better interpretation of VAT zero and SEK

  • Improved accounting prediction


  • Fix regarding not writing over interpreted Currency and zero VAT value with previous value

  • Big improvements on predicting/setting debit and credit for accounting

  • Improvements regarding account prediction

  • Extractor speed improvements

  • Online model improvements

  • Speed improvements regarding uploading and interpreting documents

  • General speed improvements


  • Account prediction: Optimized speed and accuracy

  • Performance improvements: Slight optimization of interpretation time, allround and in particular for OCR


  • Field formatting improvements (less false positives)

  • Updated offline interpretation classifier improvements (improved accuracy)

  • Faster interpretation

  • Improvements regarding avoiding setting VAT to zero


  • Increased stability: Internal updates to make sure the service can better handle high loads (no change to the 20-document limit at the moment)

  • Account predictor upgrade: Brand new algorithm put in place which promises higher accuracy on account prediction

  • Opened up api for predefined accounts: GET and PUT for predefined accounts, which are collection-specific standard accounts for total, VAT, and rounding.


  • New interpreter help algorithm based on statistical methods (more valid documents and higher accuracy)

  • Stability improvements (more robust on managing internal and external services going down or timeout


  • Stability fixes regarding uploading items (less items that could hang in process state)

  • Retry with longer delays if storage services are not responding (avoid error state for shorter disturbances with external services)

  • Minimizes setting values with just a part of the value (reduces false positives and increased accuracy)

  • Manage more monetary formats (more valid items)

  • Manage more currency formats (more valid items)

  • Manage more date formats (more valid items)

  • Stricter validation of invoice number (less false positives and increased accuracy)

  • Minimizing setting false common values (helps minimizing setting VAT to "0" etc.)

  • Speed up of OCR and mange large documents better

  • Swagger documentation updated

  • Improvements of online trained interpreter (increased accuracy and more valid items)


  • Stricter rules for monetary fields

  • Accept more monetary formats (more valid documents)

  • Accept more date formats (more valid documents)

  • Interpreter speed improvments on larger documents

  • Accept more currency formats (more valid documents)

  • Improvements on our online and offline trained model (increased accuracy)

  • Improvements on our help algoritms to lower false positives