Skip to content
KisunoAI
Self-host for free Get started
Pricing Docs Blog Get started Self-host for free

Market comparison

AI platforms compared: what matters for European companies

Language models have become interchangeable. The difference lies in the data path, in permission control and in licence lock-in. This page compares seven solutions against six criteria taken from everyday work, with a source for every verifiable figure.

In brief

  • The market splits into three classes: US cloud services (ChatGPT, Microsoft 365 Copilot, Claude), European SaaS platforms (Langdock) and open-source platforms that you either run yourself or take as a managed service (Open WebUI, LibreChat, Kisuno).
  • Seat conditions differ widely: ChatGPT Business requires at least 2 seats, Claude Enterprise at least 20 seats, and Microsoft 365 Copilot requires a qualifying Microsoft 365 licence in addition to its own fee.
  • Storage location is where the field separates: Claude workspaces are located in the USA, because “us” is the only available workspace geo. For customers registered in the EU or EFTA, Microsoft 365 Copilot is a service within the EU Data Boundary, with several documented exceptions. OpenAI offers data residency in Europe for new Enterprise and Edu workspaces.
  • Open WebUI is published under the “Open WebUI License”, the BSD-3 text with a branding clause: above 50 users in a rolling 30-day period, the branding may only be removed with an enterprise licence or explicit written permission. LibreChat, the basis of Kisuno, is published under MIT.
  • Kisuno will shortly be published as open source under MIT, self-hostable from a single user upwards with no seat minimum, and is optionally operated in Switzerland. The managed instance costs CHF 7 (EUR 7.50) per person per month, through the end of the current month with 50 percent off on a twelve-month term. The product is available; access to the managed instance is allocated in stages because demand is high.

As of August 2026. Details based on publicly available information from the respective providers. The sources are numbered at the end of the page.

The three classes of AI platform

AI platforms for companies fall into three classes: US cloud services, European SaaS platforms and open-source software that you either run yourself or take as a managed service. The class determines the data path, the pricing model and how dependent a company remains on its provider. The choice of language model almost never determines any of that.

US cloud services

ChatGPT Business and Enterprise, Microsoft 365 Copilot and Claude are finished services with a high level of maturity, a large feature set and per-seat billing. The source code is closed, and operations lie with the provider. For all three, data residency is a matter of configuration or contract, not an architectural property.

European SaaS platforms

Langdock stands for the European SaaS class: operation in the EU, ISO 27001 and SOC 2 Type II, per-seat billing.15 The source code remains closed. Operation in your own cloud or your own data centre is available, but according to the provider only from 5,000 seats upwards.15

Open source: self-hosted or managed

Open WebUI and LibreChat are open-source chat interfaces that you run on your own infrastructure. There is no provider who sees your data, but also none who takes over operations, updates and support. Kisuno is a thin fork of LibreChat that will shortly be published under the MIT licence, and adds permission management, cost control and managed Swiss hosting as an option to this class. Kisuno therefore sits in two classes at once: open source for self-hosting and a managed European offering with a fixed price per person per month. What self-hosting demands in hardware and effort is set out on a page of its own.

Six criteria that make the difference in practice

Six questions decide the selection in practice: where the data is processed, who may access which company data, what growth costs, how large the model selection is, whether your own systems can be connected, and how expensive a change of provider becomes. Everything else is detail.

Where is the data processed?

Storage location is regulated differently across the three US services. For Claude, “us” is currently the only available workspace geo, and it can no longer be changed once the workspace has been created.12 OpenAI offers data residency for new Enterprise and Edu workspaces as well as for eligible API customers, and now additionally inference residency for the region “Europe (EEA + Switzerland)”. Business workspaces are not covered.4

For customers who have registered in the EU or in EFTA, Microsoft 365 Copilot is a service within the EU Data Boundary.7 Microsoft names several exceptions: there are limited cases in which customer data is still transferred outside the boundary, and models provided by Anthropic are currently excluded from the EU Data Boundary.87 On top of that comes “Flex Routing”: model processing may take place in the USA, in Canada or in Australia at peak load, and for tenants created after 25 March 2026 this is active by default.23

Kisuno runs in ISO 27001 certified data centres operated by Green in Lupfig, Switzerland; conversations, files and accounts are held there. Depending on the choice, the models compute in the AWS data centre in Frankfurt or at European providers on servers in Europe. AWS is a US company, so the US CLOUD Act applies even with servers in Frankfurt; at the European providers it does not. The certification is held by the data centre operator; a dedicated ISO 27001 certification of Kisuno is in progress. Anyone self-hosting determines the location entirely themselves.

One limitation no provider can argue away: web search leaves your own infrastructure. With Kisuno, search queries go to external search engines without user identity; with Microsoft 365 Copilot they go to the Bing search service.7 “Everything stays in Europe” applies to storage and inference, not to the search query itself.

Who may access which company data?

As soon as an AI is connected to a line-of-business system, the decisive question is no longer what the model can do but what the individual employee is allowed to see through it. Providers solve this at three different levels: through existing system permissions, through organisation-wide rules per connector, or through a rule chain of their own ahead of the model.

Microsoft 365 Copilot shows each user only content for which they have at least read permissions in Microsoft 365.7 That is elegant as long as all relevant data sits in Microsoft 365, and ineffective for everything outside it. With Claude, owners on Team and Enterprise plans can define organisation-wide which actions a connector may perform, for example read-only.13

Kisuno enforces permissions in the MCP gateway between chat and line-of-business system, down to the individual tool: read invoices yes, read salaries no. Evaluation follows this order: individual user, role, group, server default. Every tool call is written to the audit log and can be exported. How this rule chain works in detail is set out on the governance page.

What does it cost when the team grows?

With all commercial providers the price grows linearly with the number of employees, and with two of them there is an additional lower limit. ChatGPT Business costs 25 US dollars per user and month with monthly payment, 20 US dollars with annual payment, and requires at least 2 seats.1 According to OpenAI these figures apply in most countries; for Switzerland, OpenAI states CHF 15 per user and month with annual payment.2 The price for ChatGPT Enterprise is not published.2

Microsoft 365 Copilot costs 30 US dollars per user and month with annual payment; for smaller companies there is Copilot Business from 18 US dollars per user and month.24 Both additionally require a separate licence for a qualifying Microsoft 365 plan.6 Claude Team costs 20 US dollars per seat and month with annual payment and is aimed at teams of 2 to 150 people; Claude Enterprise requires at least 20 seats.1011 Langdock states 25 euros per standard seat and month.14

With open-source software the effort does not depend on the number of people. Kisuno will shortly be published under the MIT licence: self-hosting then incurs no licence costs, from a single user upwards, with no seat minimum and no mandatory branding. What remains are server costs and language model consumption, both visible in the admin panel per person, group and model. The managed instance costs CHF 7 (EUR 7.50) per person per month. Through the end of the current month: 50 percent off with a twelve-month term, that is CHF 3.50 (EUR 3.75) per person per month.

How many models are available?

Model selection is a lock-in topic, not a quality topic. With ChatGPT and Claude you get the models of the respective house. Microsoft 365 Copilot uses models provided by Microsoft, and administrators additionally decide on the use of third-party models.7 Langdock describes itself as model-agnostic.16

Kisuno offers the current models from OpenAI (ChatGPT) and Anthropic (Claude); they run in the AWS data centre in Frankfurt. AWS is a US company, so the US CLOUD Act applies even with servers in Frankfurt, and we say so openly. For customers with higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available, operated by European companies on servers in Europe; the US CLOUD Act does not apply there. Your own API key and your own endpoint can be entered by configuration. With Open WebUI and LibreChat you choose the providers entirely yourself, but you also configure them yourself.

Can our own ERP system be connected?

All the platforms compared can connect external systems today, and all of them use the Model Context Protocol in some form for this. ChatGPT supports custom connectors via MCP.5 With Microsoft the route runs through Copilot Studio, where MCP is generally available.9 Claude connects through connectors built on MCP.13

The difference lies not in the protocol but in the control over it. LibreChat can govern access to MCP servers via roles and via a per-server release; individual tools within a server and servers from the configuration file are not covered.2122 This is exactly where the MCP gateway of Kisuno comes in: it checks every single tool call, logs the decision and denies if the check errors out instead of letting it through.

What happens when you change provider?

Migration effort is the most frequently overlooked cost item. With closed SaaS services, conversations, prompt collections, agents and connector configurations remain in the provider's system, and pricing is entirely in their hands. With a permissively licensed open-source platform, the software remains usable even once the business relationship ends.

This is exactly where the two open-source options differ. Open WebUI is published under the “Open WebUI License”, the BSD-3 text with an additional branding clause. Above 50 users in a rolling 30-day period, the branding may only be removed or replaced if an enterprise licence or explicit written permission from the rights holder is in place; code up to and including version 0.6.5 remains unaffected under BSD-3.17 LibreChat is published under MIT, without this condition20; Kisuno will shortly appear under the same licence. Why this question matters so much to us is set out on the vision page.

The comparison table

Licence, operation and access conditions of seven AI platforms. As of August 2026, checked against the providers' public documentation. Details on Kisuno are our own statements. For sources see the footnotes.
Solution Source code and licence Operation and data storage Access conditions
Kisuno MIT licence, fork of LibreChat; source code coming soon Managed in Switzerland (Green, Lupfig), in your cloud or in your own data centre Self-hosting with no licence costs from 1 user, no seat minimum, no mandatory branding; managed instance CHF 7 (EUR 7.50) per person per month, through the end of the current month 50 percent off with a twelve-month term (CHF 3.50, EUR 3.75); testing free and not limited in time, starting credit CHF 50 (EUR 55) for model usage
ChatGPT Business and Enterprise Closed Data residency and inference residency for the region “Europe (EEA + Switzerland)” for new Enterprise and Edu workspaces as well as eligible API customers; Business workspaces not covered4 Business from 2 seats, 25 US dollars per user and month, 20 with annual payment1; Enterprise on request2
Microsoft 365 Copilot Closed EU Data Boundary for customers registered in the EU or EFTA, with documented exceptions78; model processing outside the boundary at peak load (“Flex Routing”), active by default for new tenants from 25 March 202623 30 US dollars per user and month with annual payment, Copilot Business from 18 US dollars24, in each case plus a qualifying Microsoft 365 licence6
Claude (Team, Enterprise) Closed Stored in the workspace in the USA, “us” is the only available workspace geo12 Team 20 US dollars per seat with annual payment, for teams of 2 to 15010; Enterprise from 20 seats11
Langdock Closed Operation in the EU, ISO 27001 and SOC 2 Type II15 25 euros per standard seat and month14; own cloud or on-premise from 5,000 seats15
Open WebUI Open WebUI License: BSD-3 plus branding clause17 Self-operated, you choose location and infrastructure Above 50 users in a rolling 30-day period, branding removable only with an enterprise licence or written permission1718
LibreChat Open source, MIT licence20 Self-operated, you choose location and infrastructure No user-count condition, no mandatory branding20
Models, system integration and permission control of the same seven solutions. As of August 2026, checked against the providers' public documentation. Details on Kisuno are our own statements.
Solution Model selection Integration of your own systems Permissions and traceability
Kisuno Current models from OpenAI (ChatGPT) and Anthropic (Claude) via AWS in Frankfurt; for higher data-protection requirements open frontier models such as GLM, Kimi and DeepSeek at European providers; your own API key and your own endpoint by configuration ERP, SharePoint and your own REST systems through an MCP gateway Permissions down to the individual tool, chain of user, role, group and default; audit per call; shared monthly budget with a soft block
ChatGPT Business and Enterprise Models from OpenAI Connectors, custom connectors via MCP5 Role-based permissions, connectors releasable per workspace5; no training on business data3
Microsoft 365 Copilot Models provided by Microsoft, third-party models subject to admin decision7 Graph connectors and agents; MCP via Copilot Studio9 Access follows the existing Microsoft 365 permissions; retention and analysis via Microsoft Purview7
Claude (Team, Enterprise) Models from Anthropic Connectors based on MCP13 Organisation-wide rules per connector, for example read-only; users connect individually with their own permissions13
Langdock Model-agnostic according to the provider16 Connectors and workflows, workflows as a separate price package14 Governance as an add-on package, chargeable from 2027 according to the pricing page14
Open WebUI Any providers, configured by you MCP servers and tools, configured by you Permissions via roles and groups, additive according to the documentation; an explicit deny is not provided for19
LibreChat Any providers, configured by you MCP servers, releasable per role and per server; not per tool2122 Role and group permissions, release per server; no audit per individual tool call21

Where a statement refers only to part of an offering, this is named in the field. Providers change prices and features continuously. Check the linked original sources before making a decision.

The providers assessed individually

Each of the six solutions has a field in which it is the best choice. The following assessment names the strength first in each case, including where it speaks against Kisuno.

ChatGPT Business and Enterprise

In our assessment, ChatGPT is the most mature product in this comparison and, for many employees, the benchmark against which they measure every other interface. The brand is established, so onboarding is correspondingly short. According to its own statement, OpenAI does not train on business data from Business and Enterprise workspaces by default.3 Data residency in Europe can be selected for new Enterprise and Edu workspaces.4

The limits lie where control is required: closed source code, per-seat billing with a lower limit of 2 seats in the Business plan1 and no way to run the service on your own infrastructure.

Microsoft 365 Copilot

Microsoft 365 Copilot is built into Word, Excel, Outlook, PowerPoint and Teams and shows each user only content for which they have at least read permissions in Microsoft 365.7 No other product in this comparison reaches that depth. Anyone using Microsoft 365 across the board gets the shortest route from rollout to benefit here.

The price of that depth is lock-in. Copilot costs 30 US dollars per user and month and requires a qualifying Microsoft 365 licence.6 Outside the Microsoft world the permission logic ends, and for line-of-business systems the route runs through Copilot Studio.9

Claude

Claude is frequently preferred for long texts, analysis and code, and Anthropic published the Model Context Protocol, which all the providers compared now use. On Team and Enterprise plans, connectors can be restricted organisation-wide, for example to read-only.13

For European companies with a strict residency requirement, data storage is the sticking point: the workspace geo determines where data rests, and “us” is currently the only available option, which moreover cannot be changed after creation.12

Langdock

Langdock is the European answer in SaaS format: operation in the EU, ISO 27001 and SOC 2 Type II, model-agnostic, German-speaking sales.1516 For companies that want a finished European platform without running it themselves, this is a serious option.

Two points are relevant for smaller organisations: the standard seat costs 25 euros per month14, and operation in your own cloud or on-premise is, according to the provider, only offered from 5,000 seats upwards.15 The source code remains closed.

Open WebUI and LibreChat

Open WebUI and LibreChat are technically the strongest starting points for teams with their own IT department. Both run on your own hardware, both connect any models, and both are actively developed. LibreChat comes with access control that governs MCP servers via roles and via a per-server release; it does not cover individual tools within a server.2122

You should know about two differences. First, the licence: above 50 users in a rolling 30-day period, Open WebUI requires an enterprise licence or written permission from the rights holder if the branding is to be removed.17 LibreChat is published under MIT.20 Second, operations: updates, backups, integrations and support are on you.

Kisuno

Kisuno is a thin fork of LibreChat and will shortly be published under the MIT licence, extended by the building blocks that cause the most work when built in-house: permissions down to tool level with an audit per call, consumption control with a shared monthly budget and a soft block, roles and groups from Entra ID or maintained locally, and the current models from OpenAI and Anthropic plus open frontier models through a single point of access.

Honest about the status: Kisuno is ready for production use, and access to the managed Swiss instance is allocated in stages on request because demand is high. The brand is new, there is no multi-year reference list, and the integration depth in Word, Excel and Outlook does not match that of Microsoft 365 Copilot. The Outlook add-in is in beta and in internal use, the Excel add-in is in progress, native mobile apps are planned, and today the interface is installable as a PWA.

MIT licence Swiss operation New brand

When each solution fits

In practice, the selection follows three paths: maximum maturity and integration, European SaaS without running it yourself, or control over data and licence. Each path has a clear winner, and none of them wins all three.

When maturity and integration are decisive

  • Your company works entirely in Microsoft 365 and the AI is mainly meant to help in Word, Excel, Outlook and Teams: Microsoft 365 Copilot is the shortest route.
  • You need an established product that every employee can operate without training, and processing outside Europe is covered contractually: ChatGPT Business or Enterprise.
  • Your focus is on long documents, analysis and code, and storage in the USA has been reviewed and accepted: Claude.

When European operation matters but nobody should look after servers

  • You want a finished platform with European operation, certifications and a contract, without your own infrastructure: a European SaaS platform such as Langdock.15
  • You want European operation plus the option of taking the software over yourself later: Kisuno managed, with the MIT licence as a fallback.

When control over data and licence matters

  • You have your own IT department and want to build and run everything yourself: LibreChat or Open WebUI directly, with an eye on the branding clause of Open WebUI (see the section on changing provider).17
  • You need permissions down to tool level, an audit per call and cost control, but do not want to develop it yourself: Kisuno, self-hosted or managed.
  • You are an IT service provider and operate the platform for several customers: a permissive licence with no seat minimum and no branding obligation is the precondition for that.

When Kisuno does not fit

Kisuno is the wrong choice if your requirement is brand and ecosystem maturity rather than control. Five situations specifically speak against it:

  • You need AI directly inside the Microsoft Office applications at full depth: Microsoft 365 Copilot is further ahead there, whereas at Kisuno the Outlook add-in is in beta and the Excel add-in is in progress.
  • You expect a native iOS or Android app today: Kisuno is installable as a PWA, native apps are planned. ChatGPT, Claude and Microsoft 365 Copilot deliver native apps today.
  • You need a tender response with the provider's own ISO 27001 certification: today the certification is held by the data centre operator, and a dedicated one is in progress. Langdock states ISO 27001 and SOC 2 Type II for its own operations.15
  • You want an established provider with a long reference list: Kisuno is a young brand and allocates access to the managed instance in stages on request. In that case the three US services and Langdock are the safer answers.
  • You do not want to make decisions about roles, groups and tool permissions: governance is the core of this product, and that core requires deliberate setup.

Frequently asked questions

Which European alternative to ChatGPT is there for companies?

For companies there are two European routes: a European SaaS platform such as Langdock with operation in the EU, ISO 27001 and SOC 2 Type II, or an open-source platform that you run yourself. Kisuno belongs to the second group, will shortly be published under the MIT licence and is optionally operated as a managed service in Switzerland.

Is my data processed in Europe with Microsoft 365 Copilot?

For customers who have registered in the EU or in EFTA, Microsoft 365 Copilot is a service within the EU Data Boundary. At the same time Microsoft names several exceptions: limited cases in which data is still transferred outside, models provided by Anthropic, which are currently excluded, and “Flex Routing”, through which model processing may take place in the USA, in Canada or in Australia at peak load. For tenants created after 25 March 2026, Flex Routing is active by default.

Can I use Claude with data storage in Europe?

Not through a Claude workspace directly: the workspace geo determines where data rests, and according to Anthropic documentation “us” is currently the only available option, which cannot be changed after creation. Claude models are reachable in European regions through the platforms of the large cloud providers, which means a different contractual relationship.

What does a self-hosted AI platform cost in licence fees?

With Kisuno, nothing: the software will shortly be published under the MIT licence, self-hosting is then possible from a single user upwards, with no seat minimum and no mandatory branding. What remains are server costs and language model consumption. With Open WebUI the licence is tied to a condition: above 50 users in a rolling 30-day period, the branding may only be removed with an enterprise licence or written permission from the rights holder.

How does Kisuno differ from LibreChat?

Kisuno is a thin fork of LibreChat that will shortly be published under the same MIT licence, and adds three building blocks: the MCP gateway, which enforces permissions down to the individual tool and logs every call; cost control with a shared monthly budget and a soft block; and managed hosting in Switzerland as an option. LibreChat itself remains the basis.

Are these details binding?

No. The details on other providers come from their public pricing and product documentation, checked in August 2026, and are linked to their source below. Prices, seat conditions and features change continuously. Check the original source of the respective provider before making a decision.

Sources

All verifiable details about other providers on this page come from their public documentation, checked in August 2026. Details on Kisuno are our own statements.

  1. OpenAI Help Center, Managing billing and seats in ChatGPT Business: help.openai.com
  2. OpenAI, Business Pricing: openai.com/business/pricing
  3. OpenAI, Enterprise privacy: openai.com/enterprise-privacy
  4. OpenAI Help Center, Data residency and inference residency for ChatGPT: help.openai.com
  5. OpenAI Help Center, Developer mode and MCP apps in ChatGPT: help.openai.com
  6. Microsoft, Microsoft 365 Copilot Plans and Pricing (Enterprise): microsoft.com
  7. Microsoft Learn, Data, Privacy, and Security for Microsoft 365 Copilot: learn.microsoft.com
  8. Microsoft Learn, What is the EU Data Boundary: learn.microsoft.com
  9. Microsoft Learn, Extend your agent with Model Context Protocol (Copilot Studio): learn.microsoft.com
  10. Claude, Plans and Pricing: claude.com/pricing
  11. Claude Help Center, Purchase and manage seats on Enterprise plans: support.claude.com
  12. Claude Platform Docs, Data residency: platform.claude.com
  13. Claude Help Center, Use connectors to extend Claude's capabilities: support.claude.com
  14. Langdock, Pricing: langdock.com/pricing
  15. Langdock, Security: langdock.com/security
  16. Langdock, home page: langdock.com
  17. Open WebUI, License: docs.openwebui.com/license
  18. Open WebUI, Enterprise: docs.openwebui.com/enterprise
  19. Open WebUI, Permissions: docs.openwebui.com
  20. LibreChat, LICENSE (MIT): github.com/danny-avila/LibreChat
  21. LibreChat, Access Control: librechat.ai/docs
  22. LibreChat, MCP: librechat.ai/docs
  23. Microsoft Learn, Flex routing for Microsoft 365 Copilot: learn.microsoft.com
  24. Microsoft, Microsoft 365 Copilot Plans and Pricing (Business): microsoft.com

Finish comparing, then test

Kisuno is ready for production use. Testing is free and not limited in time; every new account starts with a starting credit of CHF 50 (EUR 55) for model usage. Because demand is high, we allocate access to the managed Swiss instance in stages; a request puts you in the next round.

Get started Self-host

Last reviewed: