This page is machine-translated from French. Read the French original.
Design principles
Five biases explain most of WivenLLM's choices. Understanding them helps to understand why the application behaves the way it does.
1. Local by default, cloud by choice
No API key is required during installation. WivenLLM starts with:
- an integrated embeddings engine that runs on the machine; it is downloaded once, on the first document import;
- a local vector database stored in a folder within the installation;
- a local database for spaces, conversations and settings.
The language model is the only element to choose, and it too can be local. Once the models are downloaded, an installation can therefore function without an outgoing connection for your messages and documents. On the workstation, the application also checks for updates, except in Strict sovereign mode. Anonymous telemetry remains disabled until activated.
Connecting to a cloud provider remains an option, and sometimes it's the right choice—for quality, speed, or a specific need. But it's an explicit decision, made by an administrator, not a default setting.
A practical consequence: the Privacy and Data page, at the bottom of the settings menu, summarizes the template provider, the embedding engine, and the vector database, and what each does with your data. A badge in the sidebar indicates whether the template is local or external. Actual outgoing connections can be viewed in the Wiven Console → Sovereign Mode: by default (Monitoring), they are logged but not blocked.
2. No component is mandatory
The language model, the embedding engine, and the vector database are three interchangeable components. WivenLLM communicates with each through a common interface: adding a provider does not change the rest of the application.
This is what allows us to:
- start locally, then move to a more powerful model without reindexing your spaces (as long as the embeddings engine does not change);
- use a different model per workspace — a fast model for routine questions, a heavier model for analysis;
- replace the local vector database with an enterprise database when the volume justifies it, at the cost of re-indexing: the change removes the documents from all spaces, and they must be added again.
3. An answer must be verifiable
Each answer built from your documents displays the source excerpts used to produce it: document title and excerpt, along with its similarity score. The citation does not indicate the page or section.
This principle is also found elsewhere:
- The event log retains the structuring actions of the instance, and the audit log (Wiven Console) keeps a linked, exportable copy, which makes any modification visible;
- The conversation history (Settings → Security → Workspace Chat) can be viewed and exported by an administrator or manager;
- when an agent uses a tool, the conversation shows it;
- If you enable it (Wiven Console → Response Control, disabled by default), a note at the end of the answers indicates statements that are not supported by any passage.
The goal is not to make the model infallible, but to make its work inspectable.
4. Control is the administrator's responsibility, not the user's.
Decisions that commit the organization — which supplier, which agent skills, which content rules, who accesses what — are grouped in the administration and, in multi-user mode, invisible to users.
These users have a deliberately simple interface: workspaces, conversations, documents. They can adjust settings that concern them (appearance, language, their own memories, their Second Brain and, if they are members of an organization, their mailboxes and calendars in My Connections) without being able to modify the instance's configuration.
5. We start small
The application is usable from onboarding, with a workspace and a document. Advanced features — workflows, scheduled tasks, template router, web widget, API, and most agent skills — are disabled or empty by default and are activated when needed.
This principle explains why some functions are invisible or ineffective until they have been activated: Settings → Experimental Features, Settings → Agent → Agent Skills, Settings → Agent → Scheduled Tasks (in single-user mode only).
This also applies to protections, which should not be assumed to be active:
| Protection | Default state |
|---|---|
| PII Shield | Disabled |
| Sovereign mode (network outputs) | Monitoring: records, does not block anything |
| Content control | Disabled |
| Reporting unsourced claims | Disabled |
| Approved models | Not taxed |
What these principles mean for you
| If you want… | The product directs you to… |
|---|---|
| Don't let anything out | Local native model + local vector base + native embeddings + strict sovereign mode |
| The best possible response quality | An external supplier, chosen and explicitly documented |
| Both, depending on the case | One template per workspace, or a template router |
| Prove what happened | Source citations + event log + audit log |
| Regulating practices | Roles (WivenLLM Business license) + content control, to be activated |
