AI Agents & Workflow Automation

Finding information, transferring data, following up on requests: routine work takes up your team’s time every day. We build AI solutions that take on these tasks inside your applications, with clear responsibilities and approval steps.

AIgent combines these components as a sovereign AI agent platform: agents, tested skills, company knowledge, model choice and controlled operation in the Swiss cloud or on-premise. The case study explains the architecture behind it.

AI agents that work inside your systems

An agent should move work forward: gather information, use tools and carry out the next steps. We connect AI agents to your APIs, data sources and applications so they can take a request through to completion.

For example: classify an incoming request, retrieve relevant customer data, prepare a reply and update the record in your system once approved.

The right harness for your workflow

An agent harness controls how an agent works: which tools it can use, what context it receives and when a person needs to decide. We integrate existing harnesses or build a solution around your requirements, including permissions, error handling and traceable execution.

Your process knowledge as reusable skills

We turn your workflows into skills: specific instructions, rules and tools for recurring tasks. Agents can use them to review documents, prepare reports or handle internal requests according to your requirements. We develop, test and maintain these skills with your team.

Put company knowledge and visual data to work

With LLMs and RAG, we make information from your documents and knowledge bases available for search, assistance and automated workflows. Where images or video are part of the process, we add vision AI for recognition, classification and analysis.

Start with one specific workflow

Suitable starting points include handling incoming requests, checking documents or preparing internal reports. We first establish whether AI is needed or a conventional integration already solves the task reliably. We assess value through processing time, errors and the effort required for human review.

Together, we choose a process worth automating. We agree on data access, success criteria and approvals, implement an initial workflow and test it against real cases. Once it proves useful, we take it into production and improve it over time.

Let’s talk about your workflow.

Frequently asked

Can we keep our existing systems and AI tools?

Yes. We review your existing APIs, models and harnesses and build on them where they meet your requirements. We develop the missing parts specifically for your workflow.

How do we stay in control of the agents?

We define which data and tools an agent can access and which actions require approval. Tests with typical and invalid inputs, execution logs and monitoring help you verify results and identify problems in production.

Can an AI agent comply with the revised Swiss Data Protection Act?

Yes, if the architecture supports it: an audit log under Art. 4 DPO, human-in-the-loop gates for sensitive actions, and clarity about where personal data actually ends up. The four requirements the revised act places on an AI system are set out provision by provision in A revDSG-compliant AI architecture.

Can we run an LLM ourselves in Switzerland?

Yes. For sensitive data we run open-weight models on Swiss infrastructure, with GPU scheduling on Kubernetes. A provider comparison, the cost arithmetic, and the points at which sovereignty breaks are in Sovereign AI inference in Switzerland.

Which models do you work with?

Whichever model fits the task and your data protection requirements: Claude, OpenAI and Gemini through their APIs, or self-hosted open-weight models where data must not leave the building. We build the architecture so the model stays replaceable.

How do you secure quality when AI writes the code?

If you want to build software with AI, our AI-assisted software development service explains the path from requirements through quality assurance to release.

Through architecture decisions and reviews by people. AI multiplies the amount of code; our job is to make sure that produces less code, not more. How we do that in practice is in AI writes the code. Who writes the architecture?