Upload documents to knowledge

Before you start

You need IAM.Docs access and documents approved for the intended workspace.

Click path

1. Sign in to the application.
2. Click Загрузка in the header or the chat attachment control Загрузить документы.
3. Choose an analysis scenario if one is required.
4. Choose automatic, manual or hybrid classification.
5. Click Выберите файлы or drop files into the panel. The panel displays formats and a 50 MB per-file limit.
6. Click Начать обработку, open Очередь and wait for completion. Batch processing can be sequential.
7. Open Каталог документов and check the document state before querying it.

Enlarge screenshot

1

1. Начать обработку

Verify and recover

File transfer is different from completed processing. Check catalog readiness. Select files before starting; split oversized files. Inspect the first processing attempt before uploading a duplicate.

Check PDF identifiers and process an existing document again

When a PDF contains source references or UUIDs, compare each complete identifier with the original after processing finishes. A line break inside a UUID must preserve its separator hyphen. A short search excerpt may omit a footnote; its absence from that excerpt alone does not prove that parser output lost the text.

The UUID line-wrap correction is being qualified on staging; this does not mean production has been updated. Previously processed documents use the corrected parser only after processing again; do not upload a duplicate. With processing permission, open Очередь, find your completed document, select Full restart from scratch (all steps including OCR will run again) and confirm with Retry processing (Russian: Полный перезапуск с начала (все этапы включая OCR будут выполнены заново) and Повторить обработку). If the document is absent from the queue or the action is unavailable, give its identifier to your workspace administrator and request processing of the existing document.

An ordinary retry that keeps OCR results does not apply the corrected parser to saved extracted text. A full restart runs OCR and the other processing stages again. Wait for completion and check the extracted text and source references again. Reprocessing can change extracted text, search excerpts and analysis results; a completed status alone does not verify an agent-generated document. An identifier already damaged in the original PDF cannot be reliably reconstructed automatically.

In the staging candidate, starting, cancelling and retrying processing require processing permission and access to the document in the current workspace. A not-found response does not grant access to another workspace’s document; check the selected workspace and ask its administrator for the required access.