Cognitive CDE - FAQs

A1. No. Admins cannot see other users' individual AI conversations. The conversation view provides a summary of the Project and does not display private user-level conversations.

A2. Yes. Deep Dive answers are based only on the selected or uploaded file. The AI is restricted to using that file as its source when generating answers.

A1. Currently, the system prompt is set on a per-user basis. This may change in a future release if a more centralised setup is introduced.

A2. No. The system prompt works more consistently in Deep Dive. In Wide Search, the system prompt may not always be followed in the same way, depending on the model and search behaviour.

A1. Vectorising a folder means making the folder content available for AI search. Files stored in the CDE can be opened, downloaded, and shared by users, but AI needs the content to be prepared in a different way before it can search and use it efficiently.

During vectorisation, file content is broken down into smaller sections and converted into searchable data. This allows Cognitive CDE to retrieve relevant information quickly when answering questions.

In simple terms, vectorisation syncs selected folder content with AI so that the AI can use that content to support answers.

A2. Admins can vectorise folders they have access to. Vectorisation is not personal to the admin who performs the action. Once a folder is vectorised, the content becomes available for AI search to users who already have permission to access that folder in the CDE.

Access still follows normal CDE permissions. Users will only receive AI answers based on content they are allowed to view.

A3. No. Vectorisation is currently static. If a folder is vectorised and a new file is added later, the new file will not automatically become available to Cognitive CDE.

Automatic vectorisation of newly added files is planned for a future release.

A4. Everyone. Vectorisation is applied across the organisation, not per admin. If a vectorised folder is removed, it is removed from AI search for all users.

Normal CDE permissions still apply, but the folder content will no longer be available to Cognitive CDE for AI responses.

A1. Wide Search uses Retrieval Augmented Generation (RAG) search across available vectorised content. It searches for relevant documents based on the user's question and uses the retrieved content to generate an answer.

Deep Dive provides more detailed analysis of a single document. It is designed to understand and reason over the selected file in more depth.

Use Wide Search to search across multiple documents. Use Deep Dive to ask detailed questions about a specific document.

A2. The 1000-page limit related to how much document content can be processed at one time. Very large files may not work as expected or may produce lower-quality answers. In some cases, the file may fail to process.

A1. Source references, such as Doc 1 and Doc 2, indicate the documents used to support the answer.

In the current beta, source references may not always link directly to the original document. This functionality is being improved.

A2. Yes. The intended behaviour is for source references to link back to the original source documents. This functionality is planned to work more like other AI tools, where users can open the source material directly from the response.

A3. Highlighted words or passages are intended to show relevant parts of the search results. However, in the current beta, highlighting may not always be consistent. This is being improved as the feature continues to develop.

A1. No token limit is currently enforced for the beta. Access is limited to selected users and customers while usage is monitored internally. A future release is expected to include a centralised data token system showing usage across users and AI agents.