AI cockpit aims at building a human control interface to AI systems. It is supposed to cover requirements from EU's AI act. More details can be found at https://www.kicockpit.eu/. All shared components necessary for this repo can be found here. There you also can find more info about all other activities & results of this project.
This software is open source and it is intended to be used a large community of software developers, deployers and operators. Usage and modifications are strongly encouraged - see contribution guide on how to get involved.
The “KI-Cockpit” (AI Cockpit) project was funded by the Federal Ministry of Labor and Social Affairs. See project repo for more details..
Cockpit is a part of a bundle of services. For deployment options see https://github.com/starwit/ai-cockpit-deployment. Cockpit is supposed to be installed on Kubernetes cluster using Helm chart. How to use Helm chart see according readme.
- For details on how to develop, build and run AI Cockpit see development instructions.
- Informations about the build pipelines can be found here.
AI Cockpit is composed by a number of components. The following diagram shows, how those components are working together. Cockpit has three main interfaces:
- Decision - here a system landscape will announce results of interpreting a problem/anomaly
- Action - Operators can approve/disapprove proposed actions
- Transparency - every module making decisions or executing actions, has to register via this interface. Registration will cover all necessary information for deployer/operators.
A more detailed look into AI Cockpit's architecture can be found here.
Interaction design and concepts to stay humans in control are described here. A quick intro can be found at Youtube.
AI Cockpit needs a data model to collect system decisions and possible actions. Data definition are described here.
Software in this repository is licensed under the AGPL-3.0 license. See license agreement for more details.