What is Syft-Flwr?
Syft-Flwr is an open-source framework that combines Flower’s federated learning capabilities with file-based communication. Train machine learning models collaboratively across distributed datasets without centralizing data—with easy setup, offline capability, and no servers required.Why Syft-Flwr?
Traditional federated learning frameworks require complex network infrastructure and continuous connectivity. Syft-Flwr takes a different approach: communication happens via file sync (Google Drive or SyftBox), enabling federated learning workflows that work offline, require no servers, and can run entirely in Google Colab notebooks.File-based communication
Train models without direct network connections—communication happens via file sync (Google Drive or SyftBox)
Zero infrastructure
No servers to maintain, no complex networking setup—just notebooks and file sync
Offline capable
Asynchronous message passing enables training even with intermittent connectivity
Privacy by design
Data never leaves its source—only model updates are shared
How it works
Syft-Flwr orchestrates federated learning through three key roles:- Data Owners (DO) - Organizations that hold private data and approve training jobs
- Data Scientist (DS) - Coordinator who proposes ML projects and aggregates results
- Flower Framework - Handles model aggregation using FedAvg or custom strategies
Data owners register datasets
Each data owner creates a dataset with mock (public) and private paths. Mock data lets data scientists develop code, while private data stays secure.
Data scientist proposes project
The DS creates a Flower project defining model architecture and training logic, then bootstraps it with
syft_flwr.bootstrap() to configure participants.Jobs submitted for approval
The DS submits training jobs to each data owner. DOs inspect the code before approving execution on their private data.
Built on Flower
Syft-Flwr is built on Flower, a robust federated learning framework. This means you get:- Standard Flower APIs - Use familiar
ClientAppandServerApppatterns - FedAvg support - Built-in federated averaging with model saving
- Custom strategies - Implement your own aggregation strategies
- All Flower features - Simulation mode, metrics aggregation, and more
Key features
Transport flexibility
Choose the transport that fits your environment:- SyftBox - Local file sync with RPC/crypto (default for local development)
- P2P - Google Drive or OneDrive sync (perfect for Colab notebooks)
Data governance
Data owners maintain complete control:- Review training code before execution
- Approve or reject jobs individually
- Data never leaves the local environment
- Mock datasets for safe code development
Developer friendly
Built for real-world workflows:- Run complete FL workflows in Google Colab (no local setup)
- Standard Flower project structure
- Python 3.12+ support
- Comprehensive logging and debugging
Use cases
Syft-Flwr is perfect for:Healthcare research
Train diagnostic models across hospitals without sharing patient data
Financial analytics
Build fraud detection models across institutions while maintaining data privacy
Edge AI
Train models across IoT devices with intermittent connectivity
Research collaboration
Enable academic collaboration on sensitive datasets
Next steps
Quickstart
Get a federated learning model running in 5 minutes
Installation
Install Syft-Flwr and set up your development environment
API Reference
Explore the complete API documentation
Examples
Learn from real-world federated learning examples
Community
Join the OpenMined community:- Slack - Join #support-syftbox for questions
- GitHub - Report bugs or contribute
- Co-Design Program - Get direct support for production use cases