Skip to main content

​
Features

​
ABBYY Phoenix Plus LLM endpoint

ABBYY Vantage can now supply the large language model itself. When the Phoenix Plus entitlement is enabled for a tenant, an ABBYY-managed LLM connection named ABBYY Phoenix Model appears under ADMIN → Configuration → Connections, and a Custom activity in a Process skill can call the model as a step in a workflow. There is no API key to obtain, no endpoint URL or model name to enter, and no HTTP, authentication, or provider-specific integration code to write. ABBYY selects and operates the model. The activity can pass OCR results and transaction data from the current transaction to the model, and returns structured JSON that later steps in the process can consume like any other transaction data. Because ABBYY owns the credential behind it, the managed connection is read-only: you can test it, but you cannot edit or delete it, and its credentials are never displayed. It works alongside your own LLM connections rather than replacing them, so existing connections to Anthropic, Google AI Studio (Gemini), Microsoft Foundry, Mistral AI, and OpenAI ChatGPT continue to work unchanged. ABBYY Phoenix Plus is available on ABBYY Vantage Cloud and requires a contracted entitlement. To enable it for your tenant, contact your ABBYY account team. Self-hosted deployments continue to use their own LLM connections. For more information, see LLMs in ABBYY Vantage.

​
Pending pickup list for finished transactions

The new GET /api/publicapi/v1/transactions/pending-pickup endpoint returns transactions that have finished processing and have not yet been deleted, so an integration can collect results without tracking which transactions it has already handled. Results are ordered by completion time with cursor-based pagination (up to 1,000 items per page) and can be filtered by status and skill. A transaction stays in the list until you delete it with DELETE /api/publicapi/v1/transactions/{transactionId}. The list uses at-least-once delivery. Every finished transaction appears at least once, but in some conditions, the same transactionId can appear more than once. Store the identifiers you have already processed and skip any that appear again, so that a repeated transaction is not processed twice. For more information, see Picking up finished transactions. (656035, 656077, 657334, 657335)

​
Fixes

​
Advanced Designer

  • Fixed an issue where the Fast Learning activity could fail to train after a skill was upgraded to Technology Core 3.0. Training now completes as expected. (677482, 678028, 678276, 678725, 679410)
  • Fixed an issue where a Fast Learning or Forms activity could stay linked to the original skill’s training and test document sets after an editable copy was created. These activities now use the editable copy’s own document sets.

​
Data Catalog

  • Fixed an issue where updating a data catalog did not clear the cached copy of that catalog, so the earlier records could still be used after the update. Data catalog updates now clear the cache.

​
Document processing

  • Fixed an issue where processing a very large document that produced a large number of business rule errors could take much longer than expected. These documents now process at the expected speed.

​
Manual Review

  • Fixed an issue in Manual Review where the Ctrl+D (Cmd+D on Mac) “go to next document” shortcut could open the browser’s bookmark dialog instead of advancing to the next document, depending on where the cursor was focused. The shortcut now reliably moves to the next document regardless of focus. (678854)

​
Monitor and reporting

  • Fixed an issue where an individual transaction could halt reporting data processing, so Quality Analytics reports stopped receiving new data. Reporting data now continues past these transactions.
  • Fixed an issue where workflow events could be dropped before reaching the reporting database, so transaction step reports were missing steps for the affected transactions. These events are now recorded reliably.

​
Documentation updates