I set up AI tools that run on your own infrastructure in Europe or in Switzerland. No SaaS subscription, no US-law access to the operation. And at every step that needs a language model, you decide where the data goes.
Book a callA notetaker joins the meeting and writes everything down. Useful, until someone asks where the recording goes. With the usual vendors the conversation ends up on the servers of a company you never signed a contract with about its content. Depending on where that vendor is incorporated, US law reaches data that physically sits in Europe.
For a standup that is tolerable. For board meetings, performance reviews, client work and negotiations it is not.
The bot joins Google Meet, Teams, kMeet and Hostpoint Meet sessions on its own, records and transcribes. Recording and transcription run on the server you own, or on one you can cancel without copies staying behind.
The transcript then does not go straight to a language model. It goes into an automation layer on the same infrastructure. That is where the summary is written, and that is where you decide which model writes it. A local model needs hardware and lets nothing out. A commercial API needs no hardware but receives the transcript. Switching between the two is a configuration change, not a migration.
If you moved off Teams or Zoom to a Swiss platform for data protection reasons, your meeting notes went with it. The well-known notetakers are built around Google Meet, Microsoft Teams and Zoom. None of them lists kMeet or Hostpoint Meet. Switching to a Swiss provider costs you exactly the feature you wanted the notetaker for.
My bot now also joins Infomaniak kMeet and Hostpoint Meet. Both run on Swiss servers, and both are open enough for an outside participant to be connected properly. No proprietary bot SDK, no partner programme you first have to be admitted to.
The chain stays in Europe end to end: the meeting runs with a Swiss provider, recording and transcription on your own server. No other third party in the data path, unless you deliberately decide to have the summary written by a commercial API.
One limitation, stated up front: if you switch on end-to-end encryption in kMeet, my bot cannot take notes. That is not a gap in the integration, it is the point of the encryption. Leave it off for sessions that are meant to be minuted, leave it on for the rest.
The bot that brought you here is this exact setup. It runs on my own server, joins my client meetings and produces the summaries I work from. Before that, a commercial service was in place.
The same chain also produces an infographic after every meeting. That is not a product feature. It is the answer to what a transcript can still trigger once it sits in your own house.
Talk this through with me →The tools are open source. I am paid for my time, not through your subscription. Which is why I will also tell you when an existing service is perfectly sufficient for your case.
The bot in your meeting was the evidence. What I set up, I have first made work on my own infrastructure and used day to day.
In the end your team runs and updates the setup without me. That is the goal, not a side effect.
No subscription, no lock-in. Each step stands on its own.
30 minutes, free. We find out whether your situation and my experience are a fit.
Two to three days at a fixed price. The result is a document with findings and prioritised measures.
You decide on the basis of a clear document – no commitment, no sales pressure.
If it fits, we carry on – as a fractional CIO mandate, guiding an AI rollout or digital transformation, or on a single project. If not, you still walk away with a clear roadmap.
Only with everyone's knowledge. Say before you start what the recording is for, where it is stored and how long it stays. A voice counts as particularly sensitive personal data, and in Switzerland secretly recording a non-public conversation is a criminal matter on top of that. How this applies to your case belongs with your legal advisers. I describe the framework; I do not assess it. In my own setup the bot handles this when it joins: what is recorded, where it is stored, that it is deleted after 30 days – and it stops if anyone objects.
That depends on whether a server already exists, how many meetings accumulate and how far the automation should reach afterwards. The first call establishes the order of magnitude before you commit to anything.
You or your existing host – the infrastructure stays on your side. In every case in Europe or in Switzerland, and with a provider that has no US parent company, so no access under US law runs through the provider. Which provider and which location I use for my own installation, I will tell you plainly when you ask.
You decide, and the decision is reversible. A local model keeps everything in-house, but it needs hardware and writes weaker summaries of long transcripts than the large commercial models. A commercial API reverses that: no hardware, better output, but a third party receives the transcript – with the large vendors, a US company. In my own installation the summary currently runs through a commercial API. That is a deliberate choice and not a contradiction: recording, transcription and storage sit on my server, and switching the model is one line of configuration.
Then everything stays: server, setup, documentation, access. The software is open source, no licence is tied to me and there is no margin for me to lose. That is precisely the point.
Through the normal meeting link, like any other participant. On joining it states what is being recorded, where it is stored and how long it stays. If a waiting room is active, someone has to let it in once, the same as any external guest.
Because they are widely used in Switzerland and because they are technically open enough to connect an external participant. On closed platforms the vendor decides who is allowed to build a bot at all. If you use a different platform, I will look at it before anything is promised.
30 minutes, free. Afterwards you will know whether the effort pays off – and if it does not, I will tell you so.
The direct route
We go through your starting point: which platform you use, kMeet or Hostpoint Meet included, which meetings this concerns and whether a server is already in place. Enough to judge the order of magnitude.
Book a callcall.bosshart.sg · 30 minutes
If a call feels too early, describe briefly what this is about. The message reaches me directly, not a shared inbox.