Install WivenLLM
Install WivenLLM on your own machine, in Switzerland
One installation, no account, no API key. Your documents and your conversations remain on your machine. And if you prefer to be accompanied, we will configure and install everything for you.
WivenLLM on your machine
Free for one user, with no time limit. No account, no API key. Download starts with a click.
This version is meant for a single person. Sharing between several users, connectors to your business software and logging start with the Business version, at CHF 70 per user, per year. The licence is purchased in a block of ten seats. View prices →
You are viewing this page on a phone. WivenLLM is installed on a computer, running Windows or macOS. Reopen this page on your computer to download it.
Download the installer
Downloaded from the browser, the installer will trigger a warning on first launch: see step 1 below.
Install from the terminal
recommendedThis is the simplest way: your system won't display any warnings because the alert is added by the browser, not by the file. Just one line to paste, nothing else.
Open PowerShell, paste this line, and press Enter. The download will start, and then the installer will open. No administrator privileges are required, and SmartScreen will not appear.
Expected SHA-256 fingerprint: · Verify:
Open Terminal, paste this line, and press Enter. The file will download and then open: drag WivenLLM into Applications. Downloaded this way, the application opens normally because macOS doesn't add the quarantine attribute that triggers the Gatekeeper block.
Expected SHA-256 fingerprint: · Verify:
Download interrupted? Add -C - to the curl line to pick it up where it stopped.
If you downloaded from the browser
The installer is not signed: your system will warn you on first launch. This is expected, and it's the primary reason for installation failure. Using the terminal command above, this warning does not appear.
Windows. SmartScreen displays "Windows protected your PC", publisher unknown. Click on More info, then on Run anyway.
macOS. Gatekeeper blocks the first opening. Make a Right-click on the application, then Open, and confirm. A regular double-click does not offer this option.
Allow thirty minutes for installation
This is measured, not estimated. The installer decompresses approximately 1.8 GB: the application, a Node environment, a Python environment, and two embedded agents.
For a good part of this time, the screen will give the impression that nothing is happening. Do not close the window and do not turn off the machine.
At the first launch
WivenLLM offers to install a local AI model. If you accept, it will download it, between 2 and 42 GB depending on the model chosen, with the progress displayed in the application.
You can decline, pause, or change the template later. Then everything runs locally, offline, without an account.
Both installers are provided without code signing. The SHA-256 hash displayed under each button is read directly from the file being served: it allows you to verify that your download is intact. View rates for a team →
Software first, model second
The application does not contain no model. It installs in one minute. On the first launch, you choose the model to run, and it downloads — only once.
Step 1
The applicationA single installation file. No account to create, no API key to enter.
Step 2
The indexing engine23 MB downloaded automatically on first launch. This is what reads your documents and makes them searchable.
Step 3
The language modelFrom 2 to 42 GB depending on the model chosen. Downloaded once, then fully usable offline.
What your machine can truly run
WivenLLM tests your configuration on first launch and only offers models that it runs correctly. For example:
| Model size | RAM recommended | Download size |
|---|---|---|
| 3 billion parameters | 6 GB | 2.0 to 2.1 GB |
| 7 to 8 billion | 12 GB | 4.4 to 4.9 GB |
| 13 to 14 billion | 16 GB | 8 to 9 GB |
| 30 to 32 billion | 32 GB | 18 to 20 GB |
| 70 billion | 64 GB | 42 GB |
RAM is the determining factor. A dedicated GPU or NPU speeds up generation, but is not mandatory. A model with 7 to 8 billion parameters is sufficient for the vast majority of document management uses.
What you can give him to read
Your files are indexed locally, on your disk. Nothing is sent anywhere.
What the desktop version does not
Five limitations, announced before installation rather than discovered afterward. They all relate to server deployment, which we install separately.
- Importing a web page by its address — the desktop version reads files, not URLs.
- Audio transcript — sound files are not yet supported.
- Dedicated business agents — invoices, contracts, HR: these agents fall under AgentHub.
- Local image generation — it requires a separate Python environment.
- External vector bases — the database remains local; the connection to an enterprise database is done via server deployment.
Do you need one of these functions? Let's talk about server deployment.
Five steps, end-to-end support
A simple and structured process, from the first contact to commissioning, without your data leaving your infrastructure.
20-minute call
20 minutes to understand your context, constraints, and practices. Together, we will identify the format best suited to your organization.
Format selection
Individual workstation, dedicated server or shared hosting model. You decide the scale and location of your data.
Custom configuration
We prepare the machine and load the open source models adapted to your needs.
Installation and delivery
Your AI is installed wherever you decide, on your workstation, your server or in a Swiss datacenter.
Getting started
We support your teams to get started and take advantage of the tool on a daily basis.
Do you prefer a machine? ready to use ?
Ready to install your private AI?
20 minutes of discussion to understand your context and guide you towards the configuration that corresponds to your reality.