Welcome

Welcome to the Federated Learning Net network and FL-Net software documentation.

Federated Learning Net is a deployed network for secure, collaborative research on sensitive data. Its goal is to make cross-institution analysis possible while keeping data local and under the strict control of the data holder. This network uses the FL-Net software stack.

FL-Net provides the software foundation for the Federated Learning Net network in a star-shaped network architecture:

• A Client encapsulates the local data and governance rules of a participating institution, supporting it with modular data-loading functionality. Many Clients can exist in one network. Clients connect only to the Platform, never to each other, and only exchange federated summary information, not raw data.
• A Platform provides discovery of relevant data and user interaction for Data Scientists over all connected Clients. Only one Platform exists in a network.

This documentation covers both the Federated Learning Net network and the FL-Net software stack, allowing you to understand the deployed Federated Learning Net network but also how another FL-Net based network could be deployed.

Terminology

• Client / Site — Client refers to the software; Site refers to the general physical or institutional place where that software runs.
• Federated Learning Net / FL-Net — Federated Learning Net refers to the deployed network; FL-Net refers to the underlying software. The network is built on the software, so, for example, a Federated Learning Net Client is automatically also an FL-Net Client. When joining any FL-Net network, you always deploy an FL-Net Client and during deployment then join the specific network, e.g. the Federated Learning Net network. The FL-Net software stack is open-source and can be used to deploy other networks as well.
• Platform — The central coordination node that Clients connect to, providing discovery and user interaction for Data Scientists. Network refers to the network as a whole, but "network" and "platform" may be used interchangeably in some contexts, since the Platform is the entry point to the network.

Who this is for

• Data holders who want to join the network by deploying their own Client while keeping local control over data, access, and approvals.
• Data Scientists who want to discover relevant datasets, request and run analyses, and inspect reproducible outputs.
• Tool developers and auditors who create, validate, and publish reusable ETL-Tools (Extract, Transform, Load), FL-Tools (Federated Learning), and Inference-Tools.
• FL-Net developers who want to contribute directly to the codebase itself.

How FL-Net works at a glance

• Data stays local at each participating institution's Client.
• The Platform provides discovery and federated learning access to the data while keeping data holders in control of what is shared and how it is used.
  • Data discovery is secured on each Client at multiple levels, such as rounding, thresholds, and minimum response time.
  • Federated runs execute only within locally approved boundaries and isolated environments, using secure and vetted tools.
• Results and artifacts are traceable for reproducibility and transparency while staying limited to the approved user group.

This separation of concerns between Clients and the Platform is what allows the Federated Learning Net network and FL-Net software to balance usability, collaboration, and governance.

Quickstart

For a Quickstart, you can jump directly to the relevant section for your role:

• Client deployment if you want to deploy your own FL-Net Client
• Client usage if you want to understand how to add and control data on your FL-Net Client
• Platform usage if you want to understand how to discover and analyze data on the FL-Net Network
• Tool development if you want to develop and contribute new tools to the FL-Net Network

However, we recommend instead the following reading order for new users, understanding Federated Learning Net and FL-Net first and then jumping to the relevant sections for your role:

1. Welcome (you are here)
2. Solutions
3. Get Started
4. Graphical overview
5. Optional: Understanding the architecture
6. One of:
   • Client deployment
   • Client usage
   • Platform usage
   • Tool development

Relationship to PoSyMed

FL-Net does not implement Workflow or Tool execution itself — instead, it builds on PoSyMed¹, a workflow execution engine. In PoSyMed, Tools are assembled into a directed acyclic graph describing a Workflow, with clear input and output mapping between Tools. Tools are instances with clearly defined Inputs, Outputs, and Hyperparameters. PoSyMed also includes a Tool and Workflow Store, along with an isolated execution engine.

FL-Net uses PoSyMed under the hood for all Workflow and Tool execution, as well as the basis for the Tool and Workflow Store, expanding it with federated capabilities and further features.

1. Süwer, Simon, et al. "Biomedical systems biology workflow orchestration and execution with PoSyMed." arXiv preprint arXiv:2604.20906 (2026).