GitLab, but not GitHub; Flux, but not Argo CD; Selenium, but not Cypress… Germany has made its choices to develop the D-Stack.
The Federal Ministry for Digital Affairs and the Modernization of the State is leading this initiative. Objective: to provide all levels of public administration with a sovereign technology platform for digital projects.
Set against this backdrop is a modernization program whose milestones were laid out last year. One strand focuses on digital affairs. Among other deadlines, it calls for identifying by June 2026 the administrative procedures well suited to AI usage. Then, by year’s end, to establish regulations aimed at avoiding duplicate data storage by public authorities. A broad rollout of a sovereign workplace is targeted for 2027.
A preference for the JavaScript ecosystem
The D-Stack (Deutschland-Stack) is meant to become tangible by 2028. The idea is to favour off-the-shelf components over bespoke development. Priority is given to open-source solutions and European offerings.
Germany does not start from scratch. It relies notably on standards and best practices stemming from its GovStack initiative. To date, it has conducted two public consultations on this front. And it has published a map of tools that could enter the D-Stack composition.
The main view – grid and cards – lists 128 technological building blocks. An ancillary tab presents them as a list… but only partially, potentially revealing preferences. Thus, in the software development portion: JavaScript, React, NextJS and Angular all appear on that list, while C++, Go, Java, PHP, Python, R, Rust and Swift are only visible in the grid/card visualization (as is TypeScript, by the way).
Angel ML, Pyro, spaCy… Older projects underpinning the D-Stack’s AI core
On the AI front, however, the presentation is uniform. It includes TensorFlow (first version in 2015) as well as PyTorch (2016), the ONNX format (2017), MLflow (2018) and Hugging Face Transformers (2018).
Among the “older” projects is Angel ML. Tencent had released the first version of this distributed machine learning platform in 2017. The project today sits under the Linux Foundation.
Another Linux Foundation project, also born in 2017: Pyro. It is a universal probabilistic programming language built on PyTorch.
Two of the tools in the current D-Stack come from German companies. On the one hand, spaCy, a Python natural language processing library. Explosion AI published its initial version in 2015. On the other hand, Haystack. Its first version was released in 2019 for this framework of AI orchestration developed by deepset.
The oldest AI project Germany references is Robot Framework (test automation in the context of RPA). Its roots go back to 2004. It is led by a Finnish-registered foundation of which Airbus and Capgemini are members.
Agentic protocols… and Microsoft
Three AI tools from the D-Stack emerged in 2023: Axolotl, a fine-tuning framework from the U.S. startup of the same name; RagFlow, an open-source RAG engine from InfiniFlow; PromptFlow, a visual tool for building AI applications, Made by Microsoft.
Four agent protocols also appear on the list. MCP is one of them. A2A as well (Google launched it in 2025 and then handed it to the Linux Foundation). The same goes for ANP (Agent Network Protocol) and AG-UI (Agent-User Interaction Protocol).
Aimed at delivering an HTTP agent, ANP appeared in 2024. It comes from a former Huawei and Alibaba engineer. The community around it has taken the lead of a W3C working group.
AG-UI is more recent (first version in April 2025). This event-driven protocol aims to standardize the connection between agents and applications exposed to the end user.
From Qdrant to piveau, Germany-made building blocks
On the data front, the ancillary tab highlights various formats (CSV, DCAT, JSON, MD, RDF, YAML, etc.)… and PostgreSQL.
PostgreSQL does not appear in the main mapping, but a bouquet of databases is listed. Specifically, Cassandra, CouchDB, HBase, MariaDB, MongoDB, MySQL, Neo4j, ScyllaDB and three vector databases:
- Chroma (from the American company of the same name, founded in 2023)
- Milvus (born in 2017 at the initiative of a Chinese-origin company; entrusted to the Linux Foundation)
- NQdrant (from the German company Qdrant)
Another Germany-made technology is the metadata catalog piveau, stemming from the Fraunhofer network. The D-Stack includes a complementary component: CKAN (Comprehensive Knowledge Archive Network). A UK-based non-profit association oversees this data storage and distribution application inspired by package-management systems.
A Kubernetes-centered arsenal
There is also German technology in the low code segment. In this case with n8n, offered by the eponymous company since 2019. Four other tools are listed:
- Appsmith (from the U.S. company of the same name; first version in 2020)
- Budibase (2019; from the Budibase company, based in Northern Ireland)
- Joget (2011; from the U.S. company of the same name)
- Node-RED (2013; originally IBM, now under the OpenJS Foundation)
No Argo CD on the deployment side, then, but Flux, CircleCI and Jenkins. Also Spinnaker (multicloud continuous delivery, released in 2025 by Netflix) and OpenKruise (management of workloads on Kubernetes; born in 2019 and currently in incubation at the CNCF).
GitHub is not named explicitly, but one of its products is: GitHub Actions.
When it comes to integration, the D-Stack clearly leans toward microservice architectures. It encompasses several proxies (Envoy, NGINX, Traefik) and several ingress controllers (Contour, Emissary), along with Istio. Portainer and Rancher are also on the list, as are HashiCorp Nomad and Docker Swarm. Kong is included for its open-source API gateway, alongside Red Hat with the community edition of OpenShift.
