> Source: https://souverana.ch/insights/souveraene-ki-plattform-en/ > Language: en # Sovereign AI platform: why it binds more than the model The model is easy to swap. The platform, the software your team uses for AI every day, holds your chats, files and knowledge, and that is what binds you longest. How to choose it sovereignly. Platform · Published 13 Jul 2026 · Updated 21 Oct 2026 · Joel Barmettler ## What the platform is in AI **The AI platform is the decision that stays longest: your team’s chats, prompts and the knowledge base built over months live in it, and the more you build there, the more expensive a switch becomes. It means the software your team uses for AI every day, from chat interface to user management. The language model underneath is easier to swap.** **In brief** - The knowledge base, meaning your processed documents from which the AI answers, lives in the platform. The more you build there, the more expensive a switch becomes. - Choose a model from OpenAI, Microsoft or Google, and you usually get their platform along with it: ChatGPT, Copilot or Gemini. A free choice of platform remains mainly with open models. - You do not have to operate everything yourself: the software, the infrastructure for it and the ongoing operations (updates, monitoring, security) can be assigned individually. - Choose a platform with an open ecosystem. Most specialised functions come from there, and only there can you inspect them and take them with you. The public debate revolves around models. In practice, the platform decides more. This is the fourth part of our series on sovereign AI. After infrastructure, it is time for the layer where your team works every day and where your data lives. **Series · Sovereign AI** 1. [What sovereign AI actually means for Swiss companies](https://souverana.ch/en/insights/was-ist-souveraene-ki/) 2. [AI strategy for SMEs: the questions to answer first](https://souverana.ch/en/insights/ki-strategie-kmu/) 3. [Sovereign AI infrastructure: why the model is the easy part](https://souverana.ch/en/insights/souveraene-ki-infrastruktur/) 4. Sovereign AI platform: why it binds more than the model (this part) 5. [Sovereign AI integration: when AI starts to act](https://souverana.ch/en/insights/souveraene-ki-integration/) ## Why the platform decides more than the model In our article on [infrastructure](https://souverana.ch/en/insights/souveraene-ki-infrastruktur/), we showed that the language model is the most easily replaceable component. With the platform, it is the other way around. It holds the chat history, the uploaded files, the prompts, the memory and, above all, the curated knowledge base your answers are drawn from. All of this builds up over months and grows with every day of use. That is why two open platforms pointed at the exact same model are still very different decisions. They differ in how they store your data, how they manage rights and users, which ecosystem they bring, and how easily you can leave again. You should check these four points before you choose a platform. ## Three roles: platform, host, operations A sound decision separates three roles that are often bundled into one offer, but can be assigned independently. The **platform** is the software. The **host** is the infrastructure on which the model and platform run. **Operations** is the responsibility for updates, monitoring and security day to day. These three roles combine freely. You can keep all three yourself, hand all three to a single provider, or mix them, for instance running an open platform through a Swiss partner on Swiss infrastructure. Choosing the software therefore does not yet decide where your data lives and who has access. Self-operated Full control, full effort Platform You The software freely selectable Hoster You The infrastructure your own hardware Operations You The ongoing operations your own team Managed Open Source You choose, a Swiss partner operates Platform You The software open source Hoster Provider The infrastructure Swiss cloud Operations Provider The ongoing operations Swiss operator Turnkey One provider, everything from a single source Platform Provider The software ready-made product Hoster Provider The infrastructure Swiss cloud Operations Provider The ongoing operations included Platform, host and operations are separable. Take them on yourself, and you keep full control but carry the full effort. Hand them to a Swiss provider, and you stay sovereign, provided jurisdiction and access are right. Most companies end up in the middle. For sovereignty, the middle is often the right choice: an open platform, hosted in Switzerland and run by a Swiss operator under Swiss law. That keeps control over the data intact, without your team having to operate the entire technology stack itself. What matters then is whom you entrust with hosting and operations, since that partner has technical access to the running system. ## Platform given or chosen Whether you can choose the platform at all depends on an earlier decision, the choice of model. Whoever commits to a model from **OpenAI, Microsoft or Google** usually receives the platform along with it: ChatGPT, [Microsoft 365 Copilot](https://souverana.ch/en/insights/microsoft-copilot-datenschutz/) or Gemini, complete with features for building your own agents (GPTs, Copilot agents, Gems). That is convenient, but it is a closed environment. Your chats, your own agents and your files live inside it, and the provider sets the rules. OpenAI itself shows that the provider decides on interfaces and formats: the older Assistants interface is being shut down in August 2026 and replaced with a new one, and organisations’ own GPTs are being moved to a new format. What you build inside a closed platform is never permanently yours. **Mistral** is the European exception: Le Chat and the Studio platform come with agents, connectors and a knowledge base, and can be run in your own data centre on request. The distinction matters: many Mistral models are open, but the platform itself is not. Open model weights do not make it an open platform. Whoever stays independent on models by choosing open weights can choose the platform freely. For general use, several mature, open-source interfaces are available: **Open WebUI** (the most feature-rich, with genuine user management and the largest ecosystem), **LibreChat** (agent-centred, with MCP and an open licence), **LobeChat** (a strong interface, but weaker administration) and **RAGFlow** (specialised in the high-quality processing of difficult documents). For more specialised tasks there are also standalone platforms, such as coding agents like **OpenCode** or **Cline**, which bring models, tools and context together. These open platforms often bring more functionality than the closed ones, but require setup and operations in return. ## The lock-in sits in the data The reason the platform choice weighs more heavily than the model choice is the volume and value of the data that accumulates inside it. A plain chat history is still easy to migrate. A knowledge base curated over months, in which your documents have been ingested, chunked, embedded and assigned rights, is not. That data also sits in the platform’s own structure, not in a neutral format. So check the exit before you choose: does the platform store data in open, exportable formats? Can you take chats, knowledge bases and configuration with you in full? With open-source platforms, the data sits in standard databases and can be extracted; with closed services, the exit depends on the provider. Sovereignty has two parts: where the data lives today, and whether you can bring it home if it ever comes to that. ## The ecosystem decides the niches A platform’s basic functions converge quickly. The difference in daily use comes from the specialised functions. They mostly come from the ecosystem, meaning the extensions and catalogues around the platform, and hardly from the core. OpenAI has its GPT Store for this, Open WebUI has server-side functions and tools plus a community catalogue, others rely on extensions and the open MCP standard, through which any internal system can be connected. For sovereignty, what matters is the nature of the catalogue. An open ecosystem can be inspected, adapted and self-hosted. A closed extension has to be taken as it comes. For a narrow, regulated, Switzerland-specific requirement, a tool you can read and host yourself is often worth more than the larger but locked catalogue. How far a platform can be opened up for your own applications belongs to the next layer, integration. ## What the platform delivers that the model does not Some capabilities are not delivered by the model; they come from the platform. Each of them calls for its own check, because each can create its own path for data to leave. **User management** is the biggest difference between platforms: roles, groups, sign-in through your corporate directory and fine-grained permissions distinguish a serious enterprise platform from a plain interface. Open WebUI and LibreChat are strong here, others weaker, such as LobeChat. **Web search** is more delicate than it looks: for the AI to search the web, the platform sends the search query to an external search service. That query can be sensitive, and it leaves your environment even when the model itself runs locally. A self-hosted metasearch tool such as SearXNG softens this, because the query goes out depersonalised and bundled. The leak cannot be avoided entirely as long as the open web is to be searched. **Code execution**, which the AI uses to analyse data for instance, requires a clear decision. Run locally, the data stays sovereign, but you are executing code generated by the model and need genuine isolation (a container or microVM). If you buy execution in as a ready-made cloud service, the provider takes over the isolation, but code and data then run at a third party. For sensitive data, the sovereign answer is to execute locally and in isolation. ## General-purpose or industry platform Most companies are well served by a general-purpose platform. For individual industries, though, there are platforms tailored to their documents, workflows and regulations, and they hold a genuine edge there. In **law**, tools such as Zurich’s **DeepJudge** or the European **Noxtua** search and understand legal holdings, complete with citations and matter logic. In **finance**, Zurich’s **Unique** covers typical workflows from research through to compliance. In **medicine**, services such as Lausanne’s **DOCumenter** write reports directly from the consultation, with processing exclusively in Switzerland. The advantage of such platforms is depth in one domain: fitting document understanding, built-in citations and review steps, ready-made specialist workflows. The price for that is twofold. First, they are mostly closed products with their own lock-in and higher licence costs. Second, and this is often overlooked: an industry platform is not automatically sovereign. Several well-known providers run [on US cloud infrastructure](https://souverana.ch/en/insights/berufsgeheimnis-cloud/), even where the data sits in Europe. Whether a specialised platform is the better choice depends on your core business. You have to check its sovereignty case by case, regardless of the marketing claims. ## The path to a sovereign platform A fully-fledged sovereign platform is more than an installed copy of Open WebUI. It takes user management and sign-in, a clean knowledge base, observability (monitoring and traceability of operations), the protection of sensitive data, and reliable operations with updates. There are three ways to get there, from the greatest control to the greatest convenience. The first is **self-operation**. You host and run the open platform entirely yourself, with maximum control and maximum effort. The second is **managed operation** through a Swiss partner. Providers such as Begasoft, Swisscom, Infomaniak, AlpineAI or kvant run a platform on Swiss infrastructure for you. Sovereignty is preserved provided the jurisdiction is right and the partner’s access to the system is regulated. The third is **preconfigured bundles**, which deliver the entire sovereign build as one package. > **6 cantons, over 100 municipalities** Begasoft builds and operates a sovereign AI platform on open models and Swiss infrastructure for the Swiss association for municipal and cantonal IT (SSGI), revFADP-compliant, with rollout from 2026. Sovereign platforms are already running in production in Switzerland. > > Source: Begasoft, SSGI mandate 2026 One example of such a bundle is the **Swiss AI Hub**, an open-source project that combines a whole sovereign platform (chat, model access, knowledge base, sign-in, observability) in one package of components and can be run in your own data centre or a Swiss one. A project like this saves much of the build-out, but is itself demanding to operate and keep current. > **Disclosure:** I helped shape the Swiss AI Hub as an architect. This is not a recommendation, and the AI Hub is not the right solution for every case. For many companies, a single Open WebUI instance or a turnkey managed service is the lighter, better-suited path. ## What this means for your company Choose the platform by four criteria: features, the ecosystem behind your specialised functions, the exit path, and who hosts and operates it. Do not let the choice of model quietly decide the platform. Want to choose platform, host and operations independently for your case, and have the sovereignty behind the marketing checked? [AI Architecture & Solution Selection](https://souverana.ch/en/leistungen/ki-architektur/) The top layer remains open: where AI starts acting inside your systems. The final part of the series covers it. ## Sources [1] Begasoft. «Begasoft, SSGI-Mandat 2026». Begasoft. 2026. Herstellerangabe, ohne URL. The author Joel Barmettler AI Architect · Souverana, Zurich Joel Barmettler guides Swiss companies from AI strategy to integration: sovereign, confidential and production-ready. He built the Swiss AI Hub as its architect and today co-owns its architecture; he personally leads every Souverana mandate. Mandates from one-person firms to Fortune 500 corporations. [Book an intro call](https://meet.brevo.com/joel-barmettler/30-minute-meeting) [More about Souverana](https://souverana.ch/en/) [LinkedIn](https://www.linkedin.com/in/joel-barmettler-b9ab361b7) ## Frequently asked questions What is an AI platform? The software your team uses for AI day to day: the interface for chat and agents, the connection to your documents (RAG, answers drawn from your own documents), user management and administration. The platform sits above the model and the infrastructure, and is the layer where your chats, files, prompts and knowledge live. Which AI platform is right for a business? It depends on the need. For most companies, a general-purpose platform, meaning one chat interface for all tasks, such as Open WebUI, LibreChat or a turnkey Swiss service, is enough. Individual industries with particular documents and regulations have specialised platforms. Beyond the features, what matters is the ecosystem, the operations model and the exit path. Why does the platform bind you more than the model? Because the model is interchangeable, but the platform holds your data: chat history, uploaded files, prompts, the memory (what the AI remembers about users) and, above all, the curated knowledge base. The more you invest in that knowledge base, the more expensive a switch becomes. That is why platforms differ from each other far more than the models underneath them. Can an AI platform be operated sovereignly in Switzerland? Yes. You can run an open platform such as Open WebUI, a widely used chat interface for AI, yourself on your own hardware, or choose a Swiss provider that hosts and operates it on Swiss infrastructure. Platform, host and operations are three separable roles here: you can take on all three yourself, hand them to a provider, or mix them. Open WebUI or an industry platform? General-purpose platforms such as Open WebUI suit most companies and can be operated sovereignly. An industry platform, for law, finance or medicine, say, pays off when your core business depends on domain-specific documents, citation requirements and regulation. An industry platform is not automatically sovereign, however; its hosting has to be checked case by case. ## Glossary - **Prompt**: The input to a language model: question, instruction and any text sent along. What is in the prompt determines the answer and the cost. - **Open weight**: Models whose weights are publicly available and may be self-hosted (e.g. Apertus, DeepSeek, Llama). A precondition for sovereign operation without vendor lock-in. - **Open source (AI)**: More than open weights: the training code and details of the training data are disclosed too, under a free licence. The Open Source Initiative published a definition for this in 2024. The Swiss Apertus model meets it; many models marketed as "open" only partially do. - **MCP**: Model Context Protocol: an open standard through which AI tools access systems and data in a controlled way, with clear permissions per tool.