This page is machine-translated from French. Read the French original.
Facility
WivenLLM can be installed in three ways. The application is the same in all three cases; what changes is who can use it and who exploits it.
Choose your mode
| Workstation | Server | Hosted instance | |
| For whom | A person | A team, a company | An organization without an available IT team |
| Multi-user | No | Yes, with the WivenLLM Business license | Yes |
| Web widget, API | Reachable from the machine alone | Yes | Yes |
| Exploitation | None | At your expense | Insured for you |
| Data | On the post | On your server | On the dedicated instance |
| Facility | Approximately fifteen minutes (extraction) | Half a day with the settings | Supplied ready to use |
Multi-user mode is what unlocks roles, invitations, organizations, and guest access. It's part of the WivenLLM Business offering, available upon request; without it, the «My Team» option and activation in the settings are unavailable. A license covers a specific number of workstations; guests do not use any of these. A workstation installation remains single-user: it only listens on that machine itself.
Some functions depend on the mode:
- Scheduled tasks only work in single-user mode;
- The ability for an agent to read and write files is only available on the workstation;
- The autonomous SEO Audit and Prospecting agents are not part of the server installation.
Workstation
A native application. It includes everything: interface, server, document extraction, database, vector index.
| System | Pack | Updates |
|---|---|---|
| Windows 10/11, 64-bit | Installer .exe | Automatic, within the application |
| Red Hat 9 / Rocky 9, x86_64, with graphical session | .rpm (or AppImage) | Assisted for .rpm files, automatic for AppImage files |
| NVIDIA DGX Spark (arm64), with graphics session | .deb (or AppImage) | Manual: reinstall the package |
| macOS | Not yet available | — |
- Download the application for your operating system.
- Install it. On Windows, installation is performed per user, without administrator privileges; the extraction takes 10 to 15 minutes, do not interrupt it. On a managed workstation, have the installation validated by your IT department and request an antivirus exclusion before installing.
- Launch it: The initial setup wizard opens. The first startup may take a few minutes.
No network configuration is required, and no ports need to be opened. Your data is stored in your user profile. To back it up, exit the application and copy its storage folder. Uninstalling the application leaves this data in place.
On Windows, if you accept it during installation, the application measures your machine, chooses a suitable local model, and downloads it on the first launch. A network connection is therefore required once, for this model and for the embeddings engine (approximately 90 MB, on the first document import).
Hardware requirements. You will need approximately 3 GB of disk space for the program, plus your documents and templates. The application runs on a standard office machine if you are using an external template provider. To run a template locally, 16 GB of RAM is recommended; a dedicated graphics card significantly improves performance. → Choosing your models
Server — native deployment
This is the documented deployment for this version: WivenLLM runs as a service on Windows Server.
Principle
- The service starts with the server and restarts any component that stops.
- The program and the data are separate. The data folder contains the configuration and its secrets, the database, documents, vectors, and models. An update only replaces the program.
- WivenLLM only listens on the server itself. Users access it in their browser through a reverse proxy (IIS or Caddy) which provides HTTPS.
Prerequisites
- Windows Server 2019 or later (or Windows 10/11).
- PowerShell launched as administrator.
- Approximately 3 GB for the program, plus data and models.
- An antivirus exclusion on the program and data folders, to be requested before installation: an antivirus can block the copy and slow down each startup.
Steps
- Prepare the server installation files provided by Wiven for the version.
- Run the installation script. It copies the program, creates the data folder and a configuration file with generated secrets, restricts access to this configuration, registers the service, and waits for WivenLLM to respond. → Instance configuration
- Run the verification script: it returns 0 when everything responds, and can therefore be used for a monitoring tool.
- Place the reverse proxy in front of the service, then open the HTTPS address in a browser and follow the initial setup wizard.
- Install the WivenLLM Business license if it is not already active, then enable multi-user mode from Settings → Security, which creates the administrator account. Next, create the accounts or send invitations; the company's Windows accounts connect in the Wiven Console → Active Directory.
Production deployment
A few points to address before granting access:
- TLS. With IIS, implement URL Rewrite and Application Request Routing, a response time of at least 600 seconds, the WebSocket protocol for agent sessions, and requests of at least 500 MB for documents and records. Only open the HTTPS port in the firewall.
- Backup. The backup script stops WivenLLM during the copy process, then restarts it, and can be scheduled. It copies the configuration, database, documents, and vectors, but not the models unless explicitly requested. Keep the backup on a separate disk, as it contains sensitive information. The restore script first backs up the existing data. Testing it is the only way to ensure it works.
- Update. A version upgrade involves rerunning the installation script with the new version: the service is stopped, the program replaced, the service restarted, and the configuration and data are preserved. Back up your data beforehand.
- Secrets. Keep them with the data. Changing them disconnects everyone and renders the saved provider keys unreadable.
- Network outputs. The provided configuration sets the sovereign mode to Strict, with no option to relax it from the console: nothing leaves the network except on the whitelist. Machines on the local network remain reachable. Plan ahead for template downloads, and allow Microsoft hosts if you connect Outlook. To modify the whitelist of a forced configuration, follow the procedure provided with the installation or contact Wiven support.
- Sizing. The built-in local engine is suitable for a single driver, not a team. For a team, connect a model server that responds in parallel on the GPU (Ollama, vLLM, or the llama.cpp server) via its OpenAI-compatible API. If the models run on the same machine, they determine the memory and GPU requirements—not the application.
- Embeddings engine. Choose it before importing anything: changing it empties the index of all workspaces.
Hosted instance
A dedicated instance is provisioned for your organization, with the same product and features. You receive a URL and an administrator account; operation, backups, and updates are guaranteed.
This is the appropriate mode when data sovereignty is required but the organization does not wish to operate the infrastructure itself.
After installation
In order:
- Initial setup wizard — template provider selection, choice between «Just me» and «My team», data management summary. The integrated embedding engine and local vector database are automatically applied, and an initial workspace is created for you.
- First steps — import a document and ask an initial question.
- Users and roles — switching to multi-user mode and creating accounts.
- Security and confidentiality — Review of network traffic, backups, access control. Activate the protections you need here: several are disabled by default. Wiven Console → Installation Health checks the startup, model, machine, documents, connections, and security all at once.
Common startup problems
| Symptom | Usual cause |
|---|---|
| The application starts but the interface remains blank | The interface cannot reach the server: check the service and the reverse proxy |
| Very long first start-up | Antivirus scan of installed files: the application waits for its server for up to 5 minutes. Wiven Console → Installation Health identifies the antivirus when it slows down startup |
| Provider error when sending a message | The wizard was not completed, or the supplier key is missing or invalid. |
| The documents are imported but remain untraceable. | They were extracted without being added to the workspace — adding them triggers indexing. |
| The responses ignore an added document | Indexing is still in progress, or the embedding engine has changed: this change removes documents from all spaces |
| A setting modified in the interface reverts to its value | On a server, a value in the configuration file overrides the interface: configure each setting in only one place. |
| A template download or Outlook connection fails on a server | Strict sovereign mode blocks it: allow the host in the whitelist, following the procedure in the Production Deployment section |
| A local model does not start | Insufficient memory for the chosen size → Choosing your models |
