Flower
Flower: A Friendly Federated AI Framework
Last verified:
What is Flower?
Flower is an open-source federated AI framework that enables researchers, engineers, and developers to build federated learning, federated evaluation, federated analytics, and fleet learning systems. The framework provides a unified approach to federate any workload, any machine learning framework, and any programming language, making it easy to bring existing ML projects into a federated setting.
Flower Intelligence is a new open-source AI platform that runs LLMs locally on-device (phones, tablets, laptops) for speed and privacy, with optional Flower Confidential Remote Compute for larger models that need extra power. This hybrid local-first AI approach delivers privacy-preserving on-device computation while seamlessly handing off to encrypted cloud compute when needed.
The platform is designed for AI researchers, ML engineers, data scientists, and enterprises across healthcare, finance, automotive, and IoT industries. With 7,000+ AI researchers and engineers in the community, 6,800+ GitHub stars, and 2,500+ dependent projects, Flower has become the industry standard for enterprise-grade federated AI.
Key features include the Flower Python SDK for easy federation of existing projects, Flower Datasets for partitioning datasets, Flower Hub for discovering community-built apps, Flower Baselines for reproducing research results, and upcoming SDKs for iOS, Android, and C++. The framework supports popular ML frameworks like PyTorch and TensorFlow, and runs LLMs including LLaMA, Mistral, SmolLM2, and DeepSeek.
Flower pricing
Pricing model: Freemium
The Flower Framework is open-source and free to use (pip install flwr). Flower Intelligence is open-source for on-device AI. Flower Confidential Remote Compute is in early access preview - developers can apply for the Flower Pilot Program to get personalized support and early access. Enterprise services and custom support are available through Flower Labs for projects from prototyping to production deployment. Talk to an expert for services ranging from 'standby support' to 'build and operate everything for you'. No specific pricing tiers published yet for Confidential Remote Compute.
Flower pros
- Open-source and free to use for the framework
- Supports any ML framework (PyTorch, TensorFlow, JAX)
- Works with any programming language
- Easy to federate existing ML projects with Python SDK
- Large active community with 7,000+ AI researchers and engineers
- 6,800+ GitHub stars and 2,500+ dependent projects
- Comprehensive tutorials and documentation available
- On-device AI for speed, privacy, and offline use
- Hybrid approach with Confidential Remote Compute for larger models
- End-to-end encryption for cloud compute
- Supports popular LLMs including LLaMA, Mistral, DeepSeek
- Flower Hub for community apps and collaboration
- Flower Datasets for reproducible federated learning
- Enterprise-grade support available from Flower Labs experts
- Used by Mozilla Thunderbird for privacy-preserving AI
- Cross-platform support (web, iOS, Android, C++)
- Federated fine-tuning capabilities on roadmap
- 180+ contributors to the project
Flower cons
- Confidential Remote Compute is still in early access preview
- Some SDKs (iOS, Android, C++) are coming soon with incomplete docs
- Large models require cloud compute, not fully on-device
- Federated learning has steep learning curve for beginners
- Requires modern devices for optimal on-device performance
- Cloud compute may have latency when offline
- Fine-tuning features still on roadmap, not yet available
- Limited model selection compared to full cloud AI services
Frequently asked questions about Flower
What is federated learning?
Federated learning is a machine learning approach that trains models across multiple decentralized devices or servers holding local data samples, without exchanging them. Flower makes federated learning accessible by allowing you to build federated AI systems that keep data on-device while training shared models, enabling privacy-preserving AI for healthcare, finance, and other sensitive domains.
How do I get started with Flower?
Get started in two steps: First, install Flower with pip install flwr. Second, create a Flower app using flwr new @flwrlabs/quickstart-tensorflow to download an app and follow instructions. Then run your app with flwr run . Flower works with your favorite ML framework to easily federate existing projects.
What is Flower Intelligence?
Flower Intelligence is an open-source AI platform to run LLMs locally in your app or remotely on Flower Confidential Remote Compute. It prioritizes on-device AI for speed, privacy, and offline use. When extra power is needed, Flower Confidential Remote Compute acts as a seamless private extension of the device without compromising privacy, security, or performance.
Which models does Flower Intelligence support?
Flower Intelligence supports LLaMA 3.2 1B/3B, LLaMA 3.1 8B, LLaMA 3.3 70B, SmolLM2 135M/360M/1.7B, DeepSeek-R1 (Distill-Llama-8B), Mistral Small 3, and Qwen 235B. Models run on-device via TypeScript or Swift, or remotely via Confidential Remote Compute. More models are coming soon.
What industries use Flower?
Flower is used across healthcare, finance, automotive, and IoT industries. In healthcare, it enables privacy-preserving AI for clinical research. In finance, it protects sensitive customer data. In automotive and IoT, it enables on-device learning for connected devices. The community deploys in every industry, region, and scale.
Is Flower open-source?
Yes, Flower is open-source. The framework has 6,800+ GitHub stars and 180+ contributors. The core Flower Framework, Flower Datasets, and Flower Intelligence are all open-source. The community welcomes contributions through the contributor documentation.
What is Flower Confidential Remote Compute?
Flower Confidential Remote Compute is a distributed cloud service that runs large AI models remotely with end-to-end encryption to protect sensitive user data. It serves as a private extension of on-device AI, automatically handling offloading when more computational power is needed. It is currently in early access preview.
How does Flower compare to other federated learning frameworks?
Flower distinguishes itself by offering a unified approach that federates any workload, any ML framework, and any programming language. Unlike other platforms, Flower supports PyTorch, TensorFlow, JAX, and more with the same API. It has the world's largest federated AI community with 7,000+ members and comprehensive tutorials by Andrew Ng's DeepLearning.AI.
What SDKs does Flower offer?
Flower offers a Python SDK (core framework), TypeScript SDK for web, Swift SDK for iOS, and upcoming SDKs for Android and C++. The Python SDK is used to easily federate existing machine learning projects. iOS and Android SDKs enable on-device training at scale, while the C++ SDK federates native on-device training.
Does Flower offer enterprise support?
Yes, Flower Labs provides enterprise-grade support unmatched in federated AI. Services range from standby support to building and operating everything for you. Experts address issues across the entire stack since they build the open-source framework and operate the platform. Talk to an expert for custom services to accelerate your federated AI project.