Where does your data live?
Applications run in your cloud, data centre or dedicated client-controlled tenant; systems of record stay yours.
Data Security & Deployment
Enterprise AI you control: your data is not our product.
Kriyanta production applications run within a client-controlled environment. Client systems remain in place and are accessed only through governed connectors.
Model useNo training
Data exposureTask only
ExecutionClient-controlled
ActionsBy policy
TraceabilityRecorded
Four questions
Answers to the questions your IT team will ask.
Only task-relevant information is sent through approved AI model routes. Where Kriyanta operates the environment, access is named, logged and revocable by the client.
What does the model see?
Task-relevant context only, sent to approved model routes under terms that exclude training. Inbound documents are treated as data, never as instructions.
What can the agent do?
Read by default. Every write-back is classified as automatic, approval-required or never.
How is it traceable?
Workflow and approval decisions are recorded, including the decision context shown to the approver.
Open-weight models
When data must not leave your infrastructure.
For particularly sensitive workflows, Kriyanta can run open-weight language models, including models from Mistral, entirely inside your own infrastructure, on-premise or in your private cloud. The model runs on hardware you control: prompts, documents and results are not sent to an external model provider. What the model sees remains limited to the task, as in every Kriyanta deployment. The model does not receive unrestricted access to your ERP, documents or databases. Each workflow controls which data is retrieved and passed to the model for a specific task, under defined access permissions.
The trade-off: Self-hosted AI requires dedicated infrastructure, operational management and a model selected for the task. Kriyanta evaluates performance, deployment requirements and cost to establish the right balance between AI capability and enterprise data control.
Deployment models
Flexible deployment, clear boundaries.
The logic sits above the stack. It runs in your environment and remains portable across model and platform. The execution logic remains independent of any single provider.
Cloud, private or on-premise
EU or US processing routes. Processing region, retention and exit terms are agreed for each deployment before go-live.
Global model ecosystem
Claude, OpenAI, Mistral and private models. Routed through EU or US endpoints to fit client requirements. Kriyanta.ai is a member of the OpenAI Partner Network.
Your systems stay the system of record
Read through your existing interfaces, such as standard APIs or an existing MCP gateway (REST · MCP · A2A). Client-controlled connectors, read-only by default, scoped to agreed systems.
Governance
Every action follows an execution policy.
Human approval gates are not a weakness. In industrial environments, they are often what makes AI deployable. The goal is structured, auditable action with a clear owner at every consequential decision point.
Automatic within policy
Rule-based steps with clear thresholds run without interruption. Every step is recorded.
Human approval required
Consequential actions are routed to the accountable person together with the full decision context.
Not permitted
What an agent must never do is explicitly excluded, regardless of model output.
Client-controlled execution · role-based access · task-limited model context · recorded decisions · no pooling, no training on your data
Works councils & co-determination
Trust belongs in the deployment model.
In Europe, trust also has a legal dimension. AI deployments that change how work is done, who does it, or how performance is monitored require works council engagement from the design phase. In Germany, introducing technical systems capable of monitoring employee behaviour or performance is subject to co-determination under Section 87(1) No. 6 of the Works Constitution Act (BetrVG).
The technology timeline and the social licence timeline are not the same. Treat Mitbestimmung as an execution variable, not a legal hurdle, and the question shifts from whether to how.
What the workflow does
Which task is automated, which data is processed and which systems are involved.
What stays with people
Personnel evaluation and hiring decisions remain with authorised HR professionals. Consequential actions require explicit approval.
What is recorded
Who has access to what, which decisions are recorded, and how long data is retained before it is deleted.
Technology ecosystem
Enterprise AI technology and execution ecosystem.
As a Databricks Built-On Partner and member of the OpenAI Partner Network, Kriyanta brings together enterprise data, advanced AI models and industrial decision logic.
Our technology capabilities include Databricks for enterprise data and AI workloads, OpenAI models for intelligent reasoning, and workflow orchestration platforms such as n8n and Make for automating and coordinating processes across SAP, ERP, CRM and other enterprise systems.
Technology selection follows the operational requirement, data-security constraints and deployment environment. The objective is measurable EBITDA improvement, working capital release and industrial value creation.
Frequently asked questions
Questions and answers
Is our data used to train AI models?
No. Context is shared only per task through approved model routes, under terms that exclude training. Client data is not pooled.
Where is our data processed?
Kriyanta production applications run in your cloud, your data centre or a dedicated client-controlled tenant. Processing can take place in the EU or the US; the region is agreed for each deployment before go-live.
Is on-premise deployment possible?
Yes. Kriyanta applications can run in the cloud, in a private cloud or on-premise. Your systems of record remain in place and are accessed only through governed connectors. Where data must not leave your infrastructure at all, open-weight language models, including models from Mistral, can run on your own hardware.
Can an AI agent change data in SAP?
Only by policy. Connectors are read-only by default. Every write-back is classified as automatic within policy, approval-required or never. The client defines the execution policy.
What happens if we change the AI model or provider?
Industrial logic, approval rights and workflow history are separated from the model and remain intact. Different tasks may use different approved models. Some tasks require no language model at all.
Does the AI model learn from our decisions?
Not in an uncontrolled way. Decisions, reasons and outcomes are captured as client-controlled records. Improvement is governed: only reviewed and approved rule changes enter the next version of the workflow. That knowledge stays inside the agreed client boundary and is not pooled across companies.
How are works councils involved?
AI deployments that change how work is done, who does it or how performance is monitored require works council engagement from the design phase. We plan for co-determination from the start. This requires a clear purpose, transparent data flows and defined human decisions.