WivenLLM — your private AI, installed on your premises, in Switzerland
Your AI works where you are.
WivenLLM installs a private artificial intelligence on your machine or in your offices. It talks, it reads your documents, and above all it runs your business processes by connecting to the tools you already use. You decide what stays with you — and you have it in writing.
Runs on your own hardware
Workstation, server in your offices or Swiss datacenter. You choose.
More than 800 skills and connectors
Your business agents plug into your existing tools, with no rebuild.
Compliance file
nLPD, professional secrecy, audit log delivered ready to use.
There is only one test of sovereignty.
Almost every “Swiss AI” on the market runs your artificial intelligence on shared servers, in a Geneva or Zurich datacenter. That is better than an American service. But your data still leaves your premises.
WivenLLM runs on your machine, in your offices, on your network.
Strict sovereignty mode, which blocks every outbound call, is available in the professional editions and is enabled workspace by workspace.
An assistant, agents, and the proof.
Three kinds of use, in a single product installed on your premises.
An assistant that knows your files
It talks, drafts, translates and summarises. It reads your documents and answers by citing its sources, file by file. When a piece of information does not exist in your corpus, it says so instead of inventing it. Workspaces partitioned by team or by mandate, personal memory per user, dictation and read-aloud, image and video generation.
Agents that get the work done
This is the difference with a chat. A WivenLLM agent reads a mailbox, extracts the data from an invoice, queries a database, produces a document, updates a record, and reports back. You decide what it is allowed to do, and it never decides on its own.
Governance that holds up in front of an auditor
Roles and permissions, organisations, single sign-on, a named event log, an exportable processing register. What your AI does is traceable, attributable and defensible.
The work, not the conversation.
Most AI tools stop at the answer. WivenLLM ships a complete agent engine: a library of more than 800 skills and connectors, a workflow builder, and an open interoperability standard to plug in your business tools.
Skills library
More than 800 skills and connectors ready to switch on: accounting, CRM, documents, e-commerce, messaging, prospecting. You enable only what you need.
Agent workflows
Chain several agents into a complete process: detection, correction, human validation, reporting.
Your business data
Database connector and access to file directories: the agent works on your real data, where it already lives.
Your everyday tools
Messaging, calendar, collaboration suite. The agent reads and prepares; you approve.
Open standard
WivenLLM speaks an open interoperability protocol: your developers and integrators can add their own connectors.
Office documents
Documents, spreadsheets, presentations and PDFs produced directly by the agent, on your machine.
Your ERP has no API? That is not an obstacle.
Our agents can also work through the existing interface of your business applications — including those that expose no programming interface at all. That is what lets us act on systems their vendors never opened, without a rebuild project.
Are your machines already able to run your AI?
WivenLLM analyses the machine it is installed on and gives it a capability score, workstation by workstation: which models it can run, at what speed, and what it can do for text generation, speech transcription, images and video. You know before you buy, not after.
Many companies already have machines capable of running a private AI. We always start by checking.
Your model belongs to you. Three years from now, too.
Online AI services update their models when they decide to, and retire the older ones on their own schedule. A workflow validated and documented in spring may stop producing the same results in autumn. For a firm that has to justify a method, that is a problem.
WivenLLM works with open, auditable models. You choose yours. We document its exact version, and we do not change it without your written agreement.
You also remain free to plug in another provider, or to run the model of your choice. Reversibility is not a contract clause, it is a property of the product.
On sovereignty, a question that is rarely asked.
Swiss hosting tells you where your bytes are. It does not tell you who owns your provider, nor which model is actually running under the hood. A provider's sovereignty can change overnight with a signature. A server in your own technical room cannot.
Built to be deployed, not improvised.
Who has access to what, and what use is made of your AI. On your infrastructure, with the same requirements as your other business applications.
- Three role levels — administrator, unit manager, user. Each one sees only what concerns them.
- Organisations and units — group your entities, subsidiaries or departments, with shared workspaces and distinct access rules. Automatic assignment by email domain.
- Single sign-on — sign in through your existing corporate directory.
- Event log — every action is timestamped and attributed by name.
A format for every kind of need.
From a single workstation to a multi-user server, WivenLLM adapts to your organisation. You choose the format; we configure the intelligence.
The price does not depend on the format: CHF 70 per user, per year in all three cases. Free for a single user on their own machine, with no sharing between users and no time limit. Support and updates are included in the subscription. See the detailed pricing →
Individual workstation
A workstation equipped and configured, with models pre-loaded. For a user who wants to start immediately, on their own.
Dedicated server
A compute unit installed on your premises or in a datacenter, as close as possible to your data. Several users at once, the power for the whole team.
Shared model
Your own instance, hosted in a Swiss datacenter and operated for you. The hardware and the costs are shared — your data never is.
The free version covers the single workstation. The Business licence is purchased in a block of ten seats, whatever the deployment format.
| Individual workstation | Dedicated server | Shared model | |
|---|---|---|---|
| Number of users | 1 user | Several users | Several users |
| Business agents | Included | Included | Included |
| Infrastructure ownership | You, on your own machine | YOU | Shared hardware, dedicated instance |
| Data location | On your workstation, in Switzerland | On your premises or in a Swiss datacenter | Swiss datacenter |
| Operational control | Total | Total | Delegated to Wiven |
| Hardware | Yours, or bought in your name. Configured by Wiven. | Yours, or bought in your name. Configured by Wiven. | Included in the rental |
| Maintenance | In your hands, or supported by us | In your hands, or supported by us | Included |
| Recommended profile | A self-sufficient user getting started | A team pooling compute power in-house | An organisation that would rather delegate hosting |
No per-request billing.
Your agents can run as much as needed: the computation happens on your own hardware. Your budget is known at signature and does not vary with usage.
An agent, a real process.
Not an assistant that answers questions: an agent that does the work and submits it to you.
Fiduciaries and accounting
An agent reads incoming invoices, extracts the data, spots inconsistencies and prepares the entry — you approve it.
Law firms and notaries
An agent compares two versions of a contract, isolates the amended clauses and drafts the summary note for the client.
Private banking and wealth management
An agent reads the analysis reports, extracts what concerns your positions and prepares the meeting write-up.
Industry and manufacturing
An agent answers your technicians’ questions on machine documentation, on the shop floor, with no connection.
Real estate and property management
An agent reads the condition reports, extracts the disputed points and drafts the letter to the tenant.
Public administration
An agent answers staff questions on internal procedures by citing the applicable rule, and drafts the replies.
Seven questions to ask any provider of private AI.
To be used with us as with anyone else. These seven answers are enough to separate the offers on the market.
| The question | Why it matters |
|---|---|
| If I cut the internet, does it still work? | Separates an AI running on your premises from an AI hosted elsewhere in Switzerland. |
| Where exactly does the processing take place — and can you prove it to me in writing every month? | A statement that there has been no leak is not proof that nothing was transferred. |
| Which model actually runs, in which version, and who owns it? | Swiss hosting says nothing about the model or its publisher. |
| Can you commit to not changing it without my agreement? | Determines whether your methods will stay reproducible and auditable. |
| Does it do the work, or does it answer questions? | A conversational assistant does not reduce your teams’ workload. |
| Does it connect to my ERP, including if it has no programming interface? | This is the question that eliminates most offers when it meets the real systems of Swiss SMEs. |
| Can my budget vary with usage? | Per-request billing makes the cost of an agent unpredictable by nature. |
What people ask us before signing.
What is WivenLLM?
A private artificial intelligence installed on your own hardware: a workstation, a server in your offices, or a dedicated instance in a Swiss datacenter. It brings together an assistant that reads your documents, an agent engine that runs your processes by connecting to your tools, and the governance features expected of a business application.
How is this different from a ChatGPT hosted in Switzerland?
Two differences. The first is where it runs: on your premises, not in a shared datacenter. The second is the nature of the product: WivenLLM is not only a conversational assistant, it is an agent engine that runs your processes by connecting to the tools you already use.
Does my data leave my company?
No, unless you have explicitly decided otherwise. Processing runs on your own hardware. Some optional features — internet search, for example — require an outbound connection: they are enabled or disabled per workspace, and every call is logged and appears in your processing register. You do not have to take our word for it: you have the count.
Do I pay per request?
No. The computation happens on your own hardware. Your budget is set at signature and does not vary with usage.
Do we have to replace our hardware?
Not necessarily. We start with an assessment of your machines: many recent ones are already capable of running a private AI. If an investment is needed, we tell you beforehand, with the figures.
Does WivenLLM work with Wiven agents?
Yes. The agents define what the AI does; WivenLLM defines where and how it runs. Both the AgentHub catalogue agents and the agents built for your business run directly on your infrastructure.
What happens if we want to leave?
Your documents, your indexes and your knowledge base stay on your infrastructure. The models are open and can be run elsewhere. Reversibility is a property of the product, not a contract clause.
We have already blocked ChatGPT with our security tool. What do you add?
A security tool prevents a use; it does not replace it. Your staff still have the same task to complete, and the block covers neither personal devices nor unmanaged browsers. WivenLLM removes the reason to work around the rule, and documents what has been processed internally.
Twenty minutes is enough to know whether it makes sense for you.
We start by understanding your processes and looking at your machines. If there is no measurable gain, we tell you.
Support, fixes and updates are included in the subscription. What is quoted separately is what we build, and an Enterprise offer is built with you when counting per user stops making sense. See what is included and the Enterprise tier →