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

What is WivenLLM

WivenLLM is a A complete artificial intelligence application, installed in your home.. It brings together in a single interface:

  • A cat connected to your documents, with source citations; ;
  • of the agents capable of performing actions in your tools; ;
  • a administration thought for an organization: users, roles, logs, usage control; ;
  • a connection layer open: local or cloud models, vector databases, business connectors, APIs.

The difference can be summed up in one sentence: The search engine, index, and documents remain where you put them.. Disconnect the internet connection; an installation configured for all-local operation will continue to function.

What WivenLLM actually does

1. He reads your documents and responds to you

You upload files to a workspace. WivenLLM converts them to text, cuts them into fragments, and calculates their vector representations (embeddings) and stores them in a vector basis.

When you ask a question, it finds the fragments closest to your question, provides them to the language model as context, and displays the answer. accompanied by the source extracts used. This mechanism is called RAG (Retrieval-Augmented Generation).

Supported formats range from PDFs and spreadsheets to emails, ebooks, images (via text recognition), and audio (via transcription). → File formats

2. He acts, he doesn't just respond

By entering @agent In a conversation, you hand over the reins to an agent: a model authorized to call SKILLS (tools). It can search the web, read and write files, query an SQL database, prepare an email, consult a calendar, produce a graph, generate an image.

You decide, workspace by workspace, which skills are available. → Agents Verification Status

3. It adapts to your organization

An administrator account configures the model providers, the vector database, users and their roles, content rules, and the application's appearance. Users, on the other hand, see a simple interface: workspaces, conversations, and documents.

What distinguishes WivenLLM

Online AI services WivenLLM
Where are your documents? On the provider's infrastructure On yours
Where does the model turn? At the supplier Wherever you decide — workstation, server, datacenter, or external provider if you choose
Works offline No Yes, in an all-local configuration
Model selection Required by the supplier Free, and modifiable per workspace
Logging Limited to what the supplier discloses Complete and local
Cost per use Charged based on token volume Useless locally; only usable if you connect to a paid internet provider.

Deployment methods

WivenLLM can be deployed in three ways, with the same application and the same features:

Workstation. A native application for Windows, macOS, and Linux. Everything is contained within the application: template, index, documents. Suitable for a single user, without network administration.

Organization server. A containerized or native installation on an internal server. This mode enables multi-user support, roles, the web widget, and the API. Suitable for a team or a company.

Hosted instance. A dedicated instance in a datacenter, operated for you. Same product, without the operating costs.

→ Facility

What WivenLLM is not

Honest documentation also indicates the limitations:

  • This is not a model of language. WivenLLM is the application that orchestrates models; the quality of the responses depends directly on the model you choose and the hardware that runs it.
  • This is not an exhaustive search engine. Document retrieval is semantic: it finds relevant passages, not an entire corpus. For questions like "list me all contracts containing X", the search parameters must be adapted. → Conversations
  • This is not a guarantee of accuracy. A language model can be wrong. Source citations exist precisely to allow for verification; they do not replace it.
  • Going completely local has a material cost. A fast, high-quality local model requires memory and, ideally, a graphics card. → Choosing your models

Next step