Apply for class 03 now!

anira: Dezentrale KI in Echtzeitanwendungen integrieren

About the project

Team Members

Valentin Ackva, Lina Campanella, Fares Schulz

License

Apache 2.0

Funding Period

In funding since 01/06/2026

What is the project about?

Those looking to integrate powerful AI features into their software typically have to rely on cloud APIs from major tech companies. This centralises development and creates a dependency on these companies. Users are forced to share sensitive data, making true privacy technically impossible. Open-source projects face an additional challenge in that they typically cannot or do not wish to outsource processing to the cloud, making them dependent on local solutions. While free AI models do exist in principle, their implementation on users' end devices often fails due to technical hurdles. anira is a software library that enables AI models to be executed locally with minimal latency. This is the only way to keep sensitive data private and ensure that applications function independently of the internet and cloud providers. During the project period, Anira will be further developed to enable image, video and text processing in addition to audio processing.

Which audience does the project address?

This is aimed at developers who face significant technical challenges when implementing local AI features and who prioritize data sovereignty. They can integrate anira directly into their project and get started right away.

What is to be achieved?

The existing library is to be expanded to include backends such as GGML, CoreML, and ExecuTorch, enabling the processing of all common AI model types. To this end, the data formats will also be adapted. In addition, computation will be extended from the CPU to the NPU and GPU to increase processing speed and efficiency. Overall, reference implementations and DevOps tools will also be added to further simplify the onboarding process.