Understand knowledge-library file uploads¶
The knowledge library holds manuals and other reference documents intended for governed AI use. Each transfer is tied to one library entry, document version and organization. This lets the system check a resumed transfer against the right file. Uploading through Records or another document flow does not add that file to the knowledge library.
This page explains file transfer, recovery and cleanup for administrators, support, managers and business readers. Upload and document-admission support are implemented; the connected knowledge-library experience, processing and publication are still under review. Continue with document acceptance and completion for final file checks and uncertain-result recovery. See the HiveAI overview for the wider assistant capabilities and their limits.
Keep the stages separate¶
| Stage | Meaning |
|---|---|
| Upload reserved | The transfer belongs to an organization, initiating user, library entry and document version, with an expected file size and expiry |
| File transferring | Verified parts of that same file are being stored and progress is recorded |
| All parts received | The recorded bytes reach the expected size; the full document still needs the completion and admission process |
| Document admitted | A later process has checked and accepted the assembled file for the library entry |
| Knowledge processed or published | Later library operations have prepared or made knowledge available under their own rules |
Reaching the end of the transfer does not by itself publish the document, make it available in answers, approve its contents or start paid processing. A progress response is not a publication receipt. The knowledge library's access and version rules still apply throughout these stages.
File and time limits¶
The dedicated transfer accepts PDF, DOCX and UTF-8 text documents, with a file-size ceiling of 25 MiB. A filename and stated file type must agree with that supported list. Creation checks file metadata; the full document still needs later content, structure and admission checks. Use the stricter completion limits, including the 180-character filename limit, from the start. Selecting an allowed format alone does not certify that a document is safe or suitable for organization-wide knowledge.
The upload has an expiry no later than 24 hours after creation, and its actual permitted window may be shorter. Use the expiry returned for that attempt. Reopening a page or repeating a creation request does not extend the saved window. Keep the original file available and unchanged while continuing a transfer.
Storage allowance is checked against current usage and existing reservations before a fresh upload is created. This is not a promise that every simultaneous request will fit. An unavailable usage check can prevent a new reservation. Transferring another copy unnecessarily can consume temporary space and complicate recovery.
Recover the same file and version¶
Stay in the knowledge-library journey for the selected organization, library entry and version. The transfer belongs to its initiating user; an upload associated with another user or version is not an interchangeable attempt. These uploads are excluded from ordinary personal document recovery, including for their original uploader.
Saved progress tells the application how much has arrived and which file part comes next. A repeated part can be accepted only when it matches the original part's size and checksum. An already-used part number cannot be replaced with changed file content. If the original file has changed, use the knowledge-library flow for the appropriate new version or attempt.
An interrupted save can leave file bytes ahead of recorded progress. Recovery checks the authoritative saved transfer before deciding whether the repeated write succeeded. If cleanup or another state change won the race, the application must refresh and reconcile the result. A conflict is not an instruction to overwrite another version or continue from a guessed byte count.
Once the upload expires or is no longer available for transfer, normal progress recovery may be unavailable. Full transfer progress also does not prove that the library has accepted the final document. Check the library entry's current result before restarting a large upload or repeating completion.
Discard and cleanup¶
An upload that has not been reserved for its final use can be abandoned through the supported knowledge-library flow. Once it is reserved, accepted for use or already removed after that use, temporary-upload discard is no longer the right action. Use the knowledge library's removal and retention rules for an accepted document.
Discard releases the temporary space reservation and attempts storage cleanup. Physical cleanup can remain queued if it fails. Retrying an already discarded upload returns that saved result; it does not necessarily repeat storage measurement or repair a later accounting failure.
Cleanup is not an all-or-nothing operation. Earlier file parts may have been removed before a later part fails verification or deletion. Support should reconcile the saved state and remaining storage; the user should not try to recover an abandoned file through an old temporary reference. See expiry and organization-data handling for the wider lifecycle boundaries.
Responsibilities and support¶
Administrators need permission to manage the selected library entry; possession of a file reference does not grant it. Support should establish the organization, library entry, version, original user, expected file, expiry and visible error before recommending another attempt. Keep credentials, temporary access references and document contents out of ordinary support messages.
Managers and executives can use these stages to identify whether a delay is in transfer, document acceptance, processing or publication. An upload count is not a count of published manuals, accepted answers or successful employee work. Storage requires encryption and checks that the saved bytes match the file, but deployment readiness and the complete journey into usable knowledge still need verification.
The private GitHub reference records the exact transfer, concurrency, cleanup and test limits. The initial reader review is still pending; no completed transfer, live provider rehearsal or audience acceptance is claimed by this documentation draft.