AI helps most where it knows your own documents: contracts, quotes, minutes, customer feedback. Yet exactly this data does not belong in public AI tools. Four in ten companies in Germany assume that employees use private AI tools anyway. Only 26 percent provide access to generative AI themselves (Bitkom, 21 October 2025, 604 companies with 20 or more employees).
At it-sa 2026 in Nuremberg, from 27 to 29 October, we are showing in public for the first time how your team can work with real company data without data protection having to say no. Behind it is SecureCloud kAI, the AI assistant for your whole team. You describe a task in your own words, and kAI gets it done with your documents, within the permissions that already apply. Recurring work is taken over by your own agent, on a schedule.
An example: "Analyse the Q3 customer feedback and send the team a presentation by Monday." kAI breaks the request down into steps. First it searches the connected sources, then it analyses the responses and builds the slides. At the end, an email draft to the team is ready.
The result comes back as a finished file, whether Excel, Word, PowerPoint or PDF, as a download right in the chat. Every answer from your documents names the document and the passage. kAI only sends the email once you approve it.
An agent in kAI works like a colleague with a job description. It has its own instructions, its own knowledge, its tools and a fixed model. You describe it in one sentence, and kAI's build assistant proposes the plan. Then you test it, publish it deliberately and share it with your team. Whoever uses it works with their own permissions.
An agent works on request or on a schedule: daily, on weekdays, weekly or monthly. The vulnerability radar, for example, looks for new vulnerabilities in FortiGate, Exchange and ESXi every weekday at eight and reports what needs doing today. The result is available on the web and, if you like, as a direct message in Microsoft Teams or Slack. Web search terms go to the search provider Brave.
Seven templates help you start, including the contract reviewer, the tender analyst and the meeting follow-up. Each one becomes your own copy that you adapt.
kAI only searches what the signed-in person may open and checks every hit against their permissions. A shared agent does not get a master key either. Every answer from your documents names its source, so it stays verifiable.
Two kinds of systems can be connected. kAI makes sources such as SecureCloud, SharePoint or Confluence searchable. New source types stay locked until your IT releases them per group. As a tool, kAI works in SecureCloud, Microsoft 365, Slack and Atlassian with Jira and Confluence, always with the personal account of the person using it. Your IT can enable further integrations such as HubSpot or GitLab.
For each chat you choose between "Ask first" and auto mode, provided your IT enables auto mode for your group. In auto mode, kAI handles steps without external effect on its own, such as looking things up, reading and calculating. A separate checking model assesses every single step.
kAI shows you an email, a meeting or a message first as a card, with recipients and the full text. It is only sent after your click, in auto mode too. We name the one exception ourselves. In auto mode, kAI may store files privately in SecureCloud and create a share link. kAI may not send that link.
kAI is not tied to a single language model. Per chat, you choose between models from Anthropic, OpenAI, Alibaba, Z.ai and DeepSeek. For every model, kAI shows who operates it and where it processes data.
The open models run at European operators in the EU. One of them, Qwen3.8 27B, is operated by IONOS in Berlin according to the provider. More models operated in Germany are in preparation. Claude by Anthropic runs in AWS EU regions, GPT by OpenAI on Microsoft Azure in the European Economic Area. For both, the product shows "EU region, US company", and your IT can switch them off per group. New model generations are available without a project on your side.
In the administration, your IT decides what applies to everyone, to groups or to individuals: models and the default model, tools down to the single action, including web search and code execution, and whether auto mode is available.
Each person has a monthly AI allowance. Usage beyond it is billed by consumption, and your IT can set limits per person or group. The audit log records events and approval decisions, without the content for approvals, and can be exported.
kAI runs in our cloud, so there is nothing for your team to install. You use the assistant in the browser, on a computer or on your phone, and as a direct message in Microsoft Teams and Slack. There you message kAI directly or a specific agent with "@". Our servers are in Germany, and tasks keep running there, even when you close your laptop.
SecureCloud kAI is available as a product of its own and as an extension for existing SecureCloud customers. For existing customers the path is especially short: their files are already with us, and kAI works with them directly.
A yes to AI comes with conditions. We go through these six points with you before the first pilot project:
The seals of SecureCloud GmbH, ISO/IEC 27001 certified (TÜV Rheinland) and BSI C5 attested, apply to our existing cloud services. kAI, as a new service, does not yet have an attestation of its own. You can find more about our evidence on our certificates page. The provider is SecureCloud GmbH from Neumarkt in der Oberpfalz, Germany. kAI is new, and we are starting with pilot projects.
Until Thursday, 29 October, you can find us at it-sa in Hall 7A, booth 7A-216. In a demo, we show you kAI on a workflow from your organisation. Request a demo.
Already using SecureCloud? Then book your kAI demo directly in our calendar for existing customers. You can find all features at a glance on the SecureCloud kAI product page.
SecureCloud kAI is the AI assistant for your whole team. It answers questions from your documents with the source, handles tasks and takes over recurring work with your own agents.
SecureCloud operates the platform in Germany. The models process data in the EU, one in Berlin according to the provider. Claude and GPT run in EU regions of AWS and Azure and can be switched off per group.
No. kAI checks every hit against the permissions of the person asking, also for shared agents. Only knowledge files you attach to an agent are visible to everyone you share it with.
No. kAI only sends emails, meetings and messages after your click, in auto mode too. There, kAI may only store files privately and create a share link.
As a tool: SecureCloud, Microsoft 365, Slack and Atlassian (Jira, Confluence), each with the person's own account. Your IT enables HubSpot, GitLab and further sources per group.
We quote the price based on scope. A monthly AI allowance per person is included, extra usage is billed, and your IT can set limits. kAI is also available as an extension for SecureCloud customers.