Five new projects strengthen NGI Pilots

We are delighted to welcome aboard five free and open source projects to contribute to NGI Taler and Fediversity. Both pilots have their own specific domain: NGI Taler is dedicated to privacy-preserving digital payments, and NGI Fediversity builds the hosting stack of the future. Each pilot has dedicated part of its budget — kindly provided by the European Commission — for supporting smaller projects that contribute to these goals. Allow us to to introduce the five new projects that will be joining the ranks of these pilot programmes.

Privacy preserving payments with GNU Taler

NGI Taler is working on an electronic payment system that offers privacy for those that make payments, while enforcing transparency on those that sell. The projects selected are both facilitating broader deployment: Taler CBOR APIs will provide resource-friendly Taler APIs with the recent IETF standard Concise Binary Object Representation (CBOR) and Cross-platform Taler wallet will — as the name indicates — create a convergent wallet for desktop and mobile.

Rock solid hosting-in-a-box

NGI Fediversity is a comprehensive effort to bring easy-to-use, hosted cloud services with service portability and personal freedom at their core to everyone. Three projects were selected: the first project will work on speeding up compliation of Nix packages using a binary AST. Another effort (Pragmatic NixOS cache) focuses on mitigating the strong dependency on a centralised cache. The FedLab project intend to enable Distributed Federated Machine Learning as a Nix-based service.

If you applied for a grant

This is the selection for the April 2026 calls for NGI Taler and NGI Fediversity. We always inform all applicants about the outcome of the review ahead of the public announcement, whether they are selected or not. If you have not heard anything, you probably applied to a later call or a different fund such as NGI Zero Commons Fund. Later calls and the other funds are still under review. You can see which call you applied to by checking the application number assigned to the project when you applied. The second number in the sequence refers to the month of the call, so for instance "04" in the case of the April (2026) call. (It should not happen, but if you did apply to one of these calls and did not hear anything, do contact us)

Meet the new projects!

(you can click or tap on the project name to fold out additional information)

Make nix-eval Faster — Ephemeral binary AST to speed up Nix evaluation

This project will try to improves the performance of the Nix package manager by reducing some redundant computations made during expressions evaluation: instead of repeatedly parsing and resolving the same .nix source files, the project explores the potential benefits of a reusable binary representation of import-resolved Nix expressions that can be loaded directly in memory while preserving the language's existing semantics and compatibility with current packages. This experiment will provide benchmarks and a practical proof-of-concept to help upstream those ideas to the existing nix-daemon implementations.

— For more details see: https://nlnet.nl/project/Nix-binary-AST
Pragmatic NixOS cache — Decentralised binary caches for Nixpkgs/NixOS

Nixpkgs is one of the largest collections of software on the planet. This constitutes a huge body of source code, which in turn requires a vast amount of compilation to arrive at usable software. In fact so much that in practice, ergonomic use of NixOS depends on a well-populated binary cache. Such a cache could easily become a single point of failure. This project works on delivering alternative solutions.

— For more details see: https://nlnet.nl/project/Pragmatic-NixOS-cache
Taler CBOR APIs — Resource-friendly Taler APIs with Concise Binary Object Representation

This project aims to complement the JSON data format used in the Taler RESTful APIs with support for CBOR, a binary data format which is inherently more resource-efficient than JSON, while being able to express JSONs full data model. By improving resource efficiency, Taler can broaden its range of application scenarios and enhance sustainability in the future. The project will include interoperability tests and an extensive evaluation to assess a multi-data format, backwards-compatible API design. Based on these insights, it will be easier to design similar components for other applications in or even outside the Taler ecosystem.

— For more details see: https://nlnet.nl/project/TALER-CBOR
FedLab — Distributed Federated Machine Learning Experimentation Service

Competing with very large infrastructure isn't easy, but scale can also be achieve horizontally having users/service providers sharing/aggregating resources. FedLab provides a turnkey federated experimentation service that anyone can deploy on their own hardware, able to run collaborative machine learning experiments across distributed nodes without centralizing sensitive data. Each FedLab instance manages local datasets, executes training runs, and shares only model updates (not raw data) with collaborating instances through a lightweight federation protocol. The service exposes a web-based dashboard for experiment configuration, progress monitoring, and result visualization. The project delivers an open-source, production-ready FedLab service packaged for NixOS, accompanying documentation, validate by a pilot deployment across nodes in the LLM+ research community.

— For more details see: https://nlnet.nl/project/FedLab
Cross-platform Taler wallet — Convergent wallet for desktop and mobile

This project will develop a fully FOSS cross-platform version of the wallet using the Qt6 framework, which can then run natively anywhere Qt6 is supported. This focuses on the Linux desktop and Ubuntu Touch. Future work could include deploying essentially the same codebase to other mobile devices, other desktops, and even bare-metal microcontrollers. The existing iOS and Android versions are used as a reference.

— For more details see: https://nlnet.nl/project/Crossplatform-Taler-wallet