KI Consulting Swiss

KI Consulting Swiss

AI that works.
In real operations.

We combine leading AI models with reliable software architecture — from coding agents and multimodal applications through to stable operations.

Start a conversation
Discover more

From model to solution.

A powerful model is only the beginning. We select the right model for each task, develop the product logic and bring the solution into reliable operation.

Where we have contributed

Experience from Swiss practice.

  • Swiss Post
  • Repower
  • KPT
  • Federal Office of Information Technology, Systems and Telecommunication FOITT
  • Kalaidos University of Applied Sciences
  • pharmaSuisse

Tools / reviewed 28 September 2026

Models, agents and local deployment.

We choose the model, agent and deployment approach based on the task, the data, the cost and the control required.

01

Codex & OpenAI

Codex for software tasks; GPT-6 Astra, Sol and Luna for different requirements. Custom agents can be built with the Responses API, Agents SDK or Agents API.

Code · Agents · APIs
02

Anthropic Claude

Claude Fable 5.1 for demanding long-running work, Opus 5.5 for complex engineering and Sonnet 5 when faster responses matter. We test the choice against the actual task.

Analysis · Engineering · Review
03

Google Gemini

Gemini 3.8 Flash for multi-step text, image and coding tasks, and 3.8 Live for real-time conversation. Nano Banana 2 and Veo 3.1 cover image and video production.

Multimodal · Voice · Image
04

Higgsfield

Higgsfield brings together image and video models from several providers. Its own tools, including Soul 2 and Cinema Studio 4.0, add options for image concepts and film scenes.

Image · Video · Production
05

Hermes Agent

This open agent from Nous Research uses tools, retains knowledge between sessions and can create reusable skills. It runs through a terminal or messaging apps, with a choice of model provider.

Skills · Tools · Sessions
06

OpenClaw

A self-hosted assistant that connects messaging apps, browsers and other tools. For these integrations, we define access and action approvals explicitly.

Integration · Automation · Approvals
07

Local LLMs

Gemma 4 and Qwen3.8 are open-weight model families that can run on your own hardware or servers. Depending on the model, Ollama, llama.cpp or vLLM may be suitable. We check memory needs, quality and data flow before choosing.

Own infrastructure · Ollama · vLLM

Selected work

Practice over promises.

These projects include apps, platforms, an online shop and interactive websites. We use AI when it solves a specific task.

For research, development, review and design, we choose models and tools to suit the task, data and operating environment.

How we work

We make AI work for you.

Our work does not end with a recommendation. Together with your team, we develop technically sound solutions, introduce them carefully and enable reliable day-to-day use.

  1. 01

    Clarify the task

    We clarify the need, available data, risks, quality target and measurable value before selecting a model.

  2. 02

    Develop the solution

    We combine models, knowledge retrieval, tools, safeguards, the user interface and existing software into a solution that can be verified.

  3. 03

    Use sustainably

    Quality measurement, cost, response time, monitoring and human control are set up so the solution can be used reliably in everyday work.

What sets our team apart.

Over 10 yearsSoftware engineering · Virtualisation · Data centres

MSc with distinctionComputer Science

DE · EN · IT · FRCollaboration in four languages

Do you have an AI initiative?

From an AI idea
to a production-ready solution.

Discuss an AI project View LinkedIn