This page is machine-translated from French. Read the French original.

Workspaces

The workspace is the central unit of WivenLLM. It includes documents, settings And conversations, and sets what the assistant can see.

What a workspace contains

The two spaces share nothing. A question asked in one system cannot access the documents in the other. This is the guarantee of product compartmentalization — the one that allows, for example, a space for each client without risk of data leakage between files.

One exception, voluntary and personal: the Second Brain, which traverses spaces but belongs only to you.

How to cut them

The correct cutting is by confidentiality perimeter and by corpus, not by topic of conversation.

Create a separate space when: - documents must not be mixed (a client, a file, an entity); - instructions differ (a legal space and a commercial space do not have the same prompt system); - authorized people are not the same; - the model must be different (a sensitive space entirely on-premises, another on an external supplier).

Stay in the same space when This is simply another topic of conversation: use a thread.

Two typical breakdowns

Per customer — One space per client, each containing only their own documents. The partitioning is structural. Suitable for trustees, firms, agencies.

By domain — a «technical documentation» area, an «HR» area, a «quality» area. Suitable for organizations with a shared but thematic corpus.

Discussion threads

In a space, each thread is an independent conversation with its own history. All threads in a space share the same documents and settings.

• New thread to change the subject without dragging out the previous context.
• Rename a thread to find it.
• Duplicate a thread to explore a variation from a given point, without losing the original.
• DELETE A thread erases its history.

Why this is important: The latest messages in a thread are forwarded to the template with each question. A thread that has drifted across three topics degrades the quality of responses and wastes context unnecessarily. Changing the topic = starting a new thread.

Space settings

Accessible via the gear icon next to the space name.

General

Name, image, and space removal. Deletion removes the documents from the index of this space and erases its conversations ; the original files remain in the document library.

Conversation

Mode, system prompt, model, temperature, history length, rejection message. → Conversations

Vector base

Number of fragments to find, similarity threshold, search mode, and space index reset.

Agent

Agent model and skills enabled for this space. → Agents Verification Status

Members — multi-user mode

Who has access to this space? A non-member user cannot see it. → Users and roles

Pinned documents

A document pin is injected in full in the context of each message in the space, without going through vector search.

Useful for: a charter, a glossary, a document template, a short procedure that the assistant should always have in front of them.

Handle with care: A pinned document consumes context with each message. Pinning a large file saturates the context window, slows down responses, increases costs with external providers, and leaves less room for passages retrieved by search.

Simple rule: Pin what is short and always relevant; let the search do the rest.

What happens during the deletion