The LLM API service marks the launch of Infomaniak’s AI as-a-Service, which will be developed over the coming months. Companies that process sensitive data can now integrate a trusted generative AI into their applications that guarantees data control. Open source, hosted in Switzerland and more competitive price-wise than ChatGPT, the AI provided by Infomaniak is able to process texts in many languages to process text and perform complex programming and data analysis tasks. On creating an account, 1M tokens are offered.

Trustworthy, open source and eco-responsible AI

With its AI-as-a-Service, Infomaniak commits to offering the best open-source AI technology at all times. The language model used at the launch of the service is based on Mixtral 8x7B, an open source technology whose algorithm and operation can be verified by engineers around the world.

This generative AI guarantees companies full data control and full compliance with Swiss and European legislation: the data exchanged with the AI is used exclusively for the customer’s needs. No information is analysed or resold, requests to the AI aren’t logged, and everything is hosted exclusively in Infomaniak’s infrastructure in Switzerland.

The API documentation allows developers to easily integrate AI into their workplaces to, for example, create virtual assistants, answer questions, summarise, classify, correct, generate, translate or perform sentiment analysis of content.

From an environmental perspective, our aim is to provide AI services that are as environmentally responsible as possible:

  • Our operations are powered exclusively by renewable energy.
  • The energy consumed by our services will be fully recycled thanks to our new data center.
  • We favour energy efficient models. Mixtral’s architecture, for example, manages 45 billion parameters in total, but uses only 12 billion per token, which dramatically reduces power consumption while delivering results that are superior to Llama 2 and ChatGPT 3.5.

Infomaniak’s AI services will evolve

AI is a data revolution. Scientific knowledge and computing power have been around for a long time. What AI’s been waiting for to wake up is the ability to aggregate huge datasets. Swan Blanc, AI developer at Infomaniak

The next step is to enable companies to connect our AI with their personal data to provide 100% personalised answers.

At the same time, we are gradually integrating AI into our entire ecosystem:

  • It will be possible to generate text for blog articles and web pages with Site Creator.
  • Infomaniak’s video hosting platform will allow subtitles to be generated automatically for videos.
  • Our kMeet video-conferencing solution will automatically generate a summary of your online meetings.
  • A speech-to-text API will soon be added to our AI services.

Processing sensitive data with AI

The basis for AI is data. Some players like OpenAI leave no doubt about the systematic analysis of data. Each additional piece of information that is given reinforces the power of the algorithm.

Proprietary AIs are a black box. Their operation is opaque and it is impossible to know where the data goes, whether it’s secure, how it is and will be used, for how long, etc. Other users could potentially bring private data into the public domain, entirely by chance or by questioning AI…

Infomaniak’s approach is clear and transparent:

  • Our AIs are based on verifiable open source models. It is easier to track the biases in their algorithms and the data on which they have been trained.
  • We don’t train models at the moment. Infomaniak only carries out their end use on a trusted infrastructure, with no logs, data analysis or data storage.
  • Our offering allows for specific adaptations with pre-prompts or restrictions in advance to prevent the use of AI for malicious purposes, such as generating spam.