Skip to main content

Known Limitations

A consolidated list of things Genie does not do today, so you don't waste time looking for them.

Document uploads​

  • PDF files only. The UI rejects Word, Excel, PowerPoint, Markdown, plain text, and images. The backend technically supports more formats but the web interface does not expose them.
  • Per-file size limit shown in the upload zone. The default is 100 MB; enterprise environments may be raised higher. Oversized files are rejected immediately.
  • Single file at a time. There is no multi-select / drag-multiple-files / batch upload in the UI.
  • No in-place file replacement. To update a document, delete it and re-upload the new version.
  • 5 documents per KB by default. This is a soft limit; contact Support to raise it. There is also a configurable cap on total pages indexed per KB (default 2,000).

Chat​

  • No "stop generating" button. Once you press Enter, the response streams to completion. If you regret a question, wait it out and start a new chat.
  • No conversation export. You can copy individual answers (see Copying), but there's no "save whole chat as PDF" option.
  • No thumbs-up / thumbs-down feedback. Send feedback via Support instead.
  • The KB selector locks after the first message. To switch knowledge bases mid-flight, click New Chat.
  • Conversation length is bounded by the model's context window, not by a fixed message count. Every turn re-sends the full prior conversation. Eventually you can hit the context-length error described in Error Messages — start a new chat when that happens.
  • Very long single messages are validated server-side. The default per-message limit is 50,000 characters, which is rarely an issue in practice.
  • No "search in chat history" field. You can browse the Chat History table but not full-text search it.
  • No chat rename. Titles are auto-generated from your first question.

Citations​

  • Citation panel shows filenames only. The current build does not embed a PDF viewer — you see the cited filename but cannot view the document inline. To read the source, ask the KB admin for the original file.
  • No image preview in the chat. Even if your KB documents contain figures that Genie indexes, they don't appear as thumbnails in the chat.

User interface​

  • English only. The interface is in English. No language picker is exposed (even though the framework supports 9 locales internally). AI answers can be in any language — that's based on your question's language.
  • No theme picker. The site picks light or dark mode from your operating system; you cannot override it from the UI.
  • No global search across KBs. Search happens only inside one KB at a time during chat.
  • No notifications other than email. No push, no in-app banner — the bell only shows incoming requests and your own outgoing requests.

Multimodality​

  • No image upload in chat. You cannot drop a screenshot or photo into the question input.
  • No transcription endpoint. Speech input uses the browser's Web Speech API — there is no server-side Whisper transcription.
  • Speech output limited to 5000 characters. Longer answers can't be read aloud.

Roles & access​

  • No "guest" or "view-once" role. Access is permanent until removed.
  • No expiry date on access. Memberships don't auto-expire.
  • Groups always get Reader role. To make someone admin, add them individually with Owner.
  • No bulk user import. Add users one at a time or as a group.
  • No in-app appeal for rejected join requests. Email the admin directly.
  • Join request approval has no admin note field. The user gets only Approved/Rejected — no reason. Send a separate email if you need to explain.

Knowledge bases​

  • No archive state. A KB is either live or deleted; no "deactivate" in between.
  • No KB-level settings beyond name, description, users, files. No prompt template editor, no retrieval tuning, no tags, no categories from the UI.
  • No transfer of KB ownership. The original creator's OID is permanent in the KB record (though anyone with Owner role can still manage).
  • No "departments" concept. What you might call a "department" (HR, HSE, Quality) is just a separate KB with the right Entra group assigned.

Administration​

  • No system-wide users page. Super Admins do not see a global users list.
  • No "all KBs" view for Super Admins. Super Admins see only the KBs they are assigned to, plus the + CREATE APPLICATION button.
  • No audit log in the UI. Audit events are stored server-side for compliance and accessible via Support.
  • No usage analytics dashboard. Conversation counts and token usage are tracked but not surfaced as charts in the UI.

Performance / quotas​

  • Rate limits per action. Roughly: 30 chats/min, 10 uploads/min, 5 KB writes/min, 60 KB reads/min, 60 default/min. Exceeding these returns HTTP 429.
  • Document processing can take minutes. Don't expect a large PDF to be queryable in seconds — wait for the document table to show completed.
  • No upload progress for processing. The progress bar only covers the upload itself, not the post-upload pipeline. Watch the document table for processing status.
  • No background job recovery. If a backend worker dies mid-processing, the job stays in processing indefinitely. Contact Support to unstick.

Privacy / data​

  • Logout is governed by Entra ID, not by Genie. Closing the browser tab after Logout is the simplest way to make sure cached state isn't reused on a shared device — token revocation and expiry are handled by your DHL Entra ID configuration.
  • Audit logs are not user-deletable. Required for compliance; contact Support for GDPR erasure requests.

What about features the auth doc / architecture diagram describes?​

Some features documented in the formal security and architecture papers are either:

  • Planned but not yet enabled in the UI (e.g., language picker, settings panel, batch upload)
  • Conceptual rather than implemented (e.g., "departments" / multi-tenancy — implemented per-KB, not per-department)
  • Server-side only (e.g., GDPR audit-trail export, large-format file upload via API)

If a feature you need from those documents isn't visible in the UI, Support can confirm whether it can be enabled, is planned, or needs a feature request.