When processing many documents, the agent's performance on local machines is challenging because of the many LLM calls. The problem is amplified because all documents are classified and reviewed first and then persisted. As a result, the system ignores previously performed classifications and reviews when processing is interrupted before being completed for all documents. See if performance can be improved, or implement functionality to submit positively reviewed classifications more effectively.
When processing many documents, the agent's performance on local machines is challenging because of the many LLM calls. The problem is amplified because all documents are classified and reviewed first and then persisted. As a result, the system ignores previously performed classifications and reviews when processing is interrupted before being completed for all documents. See if performance can be improved, or implement functionality to submit positively reviewed classifications more effectively.