Flower
Flower is an open-source federated learning framework plus the SuperGrid platform for training AI across decentralized, private data.
Flower is the default pick when you genuinely need federated training and want framework freedom — PyTorch, TensorFlow, Hugging Face, and PennyLane support plus the SuperGrid governance layer (audit logs, RBAC, private federations) is hard to match in open source. It assumes ML competence: this is not a drop-in product for teams without distributed-training experience. If you need fully managed no-code AI, look at a hosted platform instead. For privacy-preserving, multi-site model training, shortlist Flower first.
Verified 7d ago · liveness 77/100 · cite: rightaichoice.com/tools/flower
- Healthcare and life-science teams federating training across hospitals without moving patient data
- Enterprises that need privacy-preserving AI on distributed data with audit logs and RBAC
- ML researchers experimenting with federated algorithms, datasets, and baselines
- Developers building collaborative AI apps with on-device training via iOS, Android, or C++ SDKs
- Teams that need a fully managed, no-code AI platform and have no distributed-training experience
- Anyone training one model on one centralized dataset — federation adds complexity without benefit
- Beginners without an ML or distributed-systems background who need hand-holding
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Flower if you're training one model on one centralized dataset — federation adds orchestration work (clients, strategies, rounds) and credit planning for zero benefit when all your data already sits in one place.
Pro tiers are billed yearly — the €20, €50, and €200 monthly figures all assume an annual commitment, and paying monthly costs up to 18% more per month.
SuperGrid Pro at €20-50/month billed yearly fits a small ML team or research group running a handful of federations; the €200/month tier and credit scaling up to 20,000,000 suit an enterprise pilot. SuperGrid Max is quote-based for regulated buyers needing Confidential Compute. Against fully managed enterprise AI platforms, Flower is cheaper per seat but you supply the ML engineering; against free open-source federated libraries, you are paying for the governance layer, not the framework.
In short
Flower — Flower is an open-source federated learning framework plus the SuperGrid platform for training AI across decentralized, private data. Best for Healthcare and life-science teams federating training across hospitals without moving patient data, Enterprises that need privacy-preserving AI on distributed data with audit logs and RBAC, ML researchers experimenting with federated algorithms, datasets, and baselines. Free to start; paid plans from €20/mo.
What's new in Flower
Checked 7 days agoAcross the latest 5 updates: 1 launch and 4 changelog entries.
Announcing Flower 1.39
Flower 1.39 stable released, the latest in a steady cadence of framework releases through September 2026.
Announcing Flower 1.38
Flower 1.38 stable released.
Announcing Flower 1.37
Flower 1.37 stable released.
Introducing Endeavor 1.0 from Flower Labs
Endeavor 1.0 is a generalist model for reasoning, coding, and long-horizon agent work, available via Flower or private deployment.
Announcing Flower 1.36
Flower 1.36 stable released alongside the Endeavor 1.0 announcement.
What people actually say about Flower — is it worth it?
We scanned public community sources for Flower on Jul 6, 2026 and could not establish that the discussion we found is about this tool rather than something else sharing its name. Our own analysis of that scan says the posts were off-subject. Rather than publish a sentiment score built on the wrong subject, we publish nothing here and re-run the scan.
Viability Score
How well maintained and how widely used is Flower? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Flwr open-source federated learning framework
- Framework-agnostic: PyTorch, TensorFlow, Hugging Face Transformers, PennyLane
- iOS SDK for federated on-device training (preview)
- Android SDK for federated on-device training (preview)
- C++ SDK for native on-device training
- Flower Datasets for reproducible federated data partitioning
- Flower Baselines for reproducing published federated research
- SuperGrid Web UI for managing and monitoring federations
- Unlimited federations on SuperGrid Pro and Max
- Public and private federations
- Audit logs for compliance reporting
- RBAC, with advanced RBAC on SuperGrid Max
- Confidential Compute on SuperGrid Max
- Flower Hub for discovering, sharing, and verifying federated apps
- Stable FAB format for packaging federated applications
About Flower
Flower is a federated learning framework (Flwr) plus a commercial platform, SuperGrid, for what the company calls Collaborative AI — training models across data that never leaves its source. The open-source core is framework-agnostic: PyTorch, TensorFlow, Hugging Face Transformers, and PennyLane all plug in, with iOS, Android, and C++ SDKs for on-device training. Flower Datasets handles partitioned data loading and Flower Baselines reproduces published federated research, backed by a community the vendor counts at 7,000+ GitHub stars, 2,500+ ecosystem projects, and 180+ contributors. The vendor also runs free DeepLearning.AI courses with Andrew Ng on federated learning and federated LLM fine-tuning. On the commercial side, SuperGrid adds a Web UI, audit logs, RBAC, public and private federations, Verified Apps, and Confidential Compute on the Max tier. Flower Models ships sovereign and frontier models, Flower Hub publishes and shares federated apps in the stable FAB format, and Flower Intelligence does on-device AI with optional confidential remote compute. In September 2026 alone, Flower shipped framework releases 1.36 through 1.39 and launched Endeavor 1.0, a generalist model for reasoning, coding, and long-horizon agent work available through Flower or private deployment. Flower Agent and FlowerBench round out an agentic push; FlowerBench benchmarks agents on proprietary, long-horizon enterprise tasks rather than demos. Flower sits between research-grade federated libraries and fully managed enterprise AI platforms. It asks more of your team than a hosted no-code tool, but you keep framework choice and deployment control that closed platforms usually reserve for their largest contracts.
Behind the Verdict
Flower's real differentiator is that the open-source framework and the commercial platform are the same stack. You can prototype federated training locally with Flwr, package it as a FAB app, and then run it under SuperGrid with audit logs, RBAC, and public or private federations, without rewriting the orchestration. That continuity is unusual — most federated libraries stop at the library, and most enterprise platforms won't let you see the framework. Framework-agnosticism is the second strength. PyTorch, TensorFlow, Hugging Face Transformers, and PennyLane are all supported, and there are iOS, Android, and C++ SDKs for on-device training. Flower Datasets handles the awkward part of federated work — partitioning data reproducibly — and Flower Baselines lets you reproduce published results before you run your own. The vendor backs this with free DeepLearning.AI courses with Andrew Ng, which lowers the on-ramp more than the documentation alone would. The commercial layer is credit-based. SuperGrid Pro tiers listed on the pricing page run €20/month billed yearly (8,000 monthly credits), €50/month (20,000 monthly credits, scalable up to 20,000,000), and €200/month (80,000 monthly credits), all billed yearly with up to 18% savings versus monthly, and a free Flower Account with 3,000 sign-up credits. SuperGrid Max covers Verified Apps, Confidential Compute, Private Apps on Flower Hub, advanced RBAC, and SOC2 reports, priced on contact. Note the pricing page currently still carries 'placeholder' labels on three of the four Pro/Max cards, so treat the tier lineup as in flux. The agentic bet is the newest piece and the one to watch. Endeavor 1.0, announced 1 September 2026, is a generalist model for reasoning, coding, and long-horizon agent work, available via Flower or private deployment. Flower Agent is a Collaborative AI agent, and FlowerBench measures agents on secure, proprietary, long-horizon enterprise tasks. Flower has not published independent third-party benchmark results for Endeavor yet, so treat its comparative standing as unverified. Where it doesn't fit: a single model on a single centralized dataset. Federation adds orchestration complexity for no benefit there. Teams with no ML or distributed-systems background will struggle — the docs assume you know what a client, a strategy, and a round are. And if you want flat predictable spend rather than credit consumption planning, the Pro tiers require forecasting your monthly credits.
Researching Flower? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Flower actually fits — and what changes day-one when you adopt it.
Install Flwr, define the model with PyTorch, point Flower Datasets at each hospital's local partition, and run simulation locally before deploying clients on-site.
Outcome: A federated diagnostic model trains across sites while raw patient records never leave each hospital's infrastructure.
Pull a Flower Baselines configuration, run the experiment with Flower Datasets partitioning, then vary the aggregation strategy.
Outcome: Reproducible baseline numbers you can build on, without reimplementing the data partitioning or the training loop.
Package the trained app as a FAB, publish it on Flower Hub, then run it under SuperGrid with private federations, audit logs, and RBAC.
Outcome: Ongoing federated training with an access-control and audit trail that satisfies compliance reviewers.
Use Cases
- Train a global medical diagnostic model across hospitals without moving patient data off-site.
- Fine-tune a large language model on confidential enterprise documents held in separate departments.
- Build a federated recommendation system from on-device mobile user data.
- Reproduce federated learning research with Flower Baselines before running your own experiments.
- Deploy collaborative AI workloads across satellites or edge devices with intermittent connectivity.
- Package a federated app as a FAB and publish it on Flower Hub for community reuse.
- Run agentic enterprise tasks with Flower Agent and measure them against FlowerBench.
- Deploy Endeavor 1.0 privately so prompts and outputs stay inside your own infrastructure.
Models Under the Hood
as of 2026-10-11
Limitations
- Flower assumes ML and distributed-systems competence; the docs and examples move fast for anyone who hasn't run a training loop before.
- Three of the four SuperGrid pricing cards are still labelled 'placeholder', so the tier lineup and credit allowances may shift — Pro tiers currently list 8,000, 20,000, and 80,000 monthly credits, with the middle tier scalable up to 20,000,000.
- Enterprise capabilities including Confidential Compute, Verified Apps, Private Apps on Flower Hub, and SOC2 reports sit on SuperGrid Max, which is priced on contact rather than published.
- Endeavor 1.0 is presented as a frontier-class model for reasoning, coding, and long-horizon agent work, but no independent benchmark comparison against other frontier models is published in the evidence.
as of 2026-10-04
Verification history
We have re-verified Flower 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Flower tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Flower Account (Free)
$0
Ideal for
An individual developer or researcher who wants to explore Flower Hub and the open-source Flwr framework before committing to a paid federation.
What this tier adds
Free entry point: sign-up credits of 3,000, Flower Hub access, and full open-source framework use.
SuperGrid Pro (€20/mo tier)
€20/mo, billed yearly
SuperGrid Pro (€50/mo tier)
€50/mo, billed yearly
SuperGrid Pro (€200/mo tier)
€200/mo, billed yearly
SuperGrid Max
Custom
Ideal for
Regulated enterprises and government buyers who need Confidential Compute and a verified app supply chain.
What this tier adds
Adds Verified Apps, Confidential Compute, Private Apps on Flower Hub, advanced RBAC, and SOC2 reports, with custom monthly credits.
Where the pricing makes sense
The company stage and team size where Flower's pricing actually pencils out — and where peers do it cheaper.
SuperGrid Pro at €20-50/month billed yearly fits a small ML team or research group running a handful of federations; the €200/month tier and credit scaling up to 20,000,000 suit an enterprise pilot. SuperGrid Max is quote-based for regulated buyers needing Confidential Compute. Against fully managed enterprise AI platforms, Flower is cheaper per seat but you supply the ML engineering; against free open-source federated libraries, you are paying for the governance layer, not the framework.
Setup time & first value
How long it actually takes to get something useful out of Flower — broken out by persona, not the marketing-page minute.
A developer already fluent in PyTorch or TensorFlow can install Flwr and finish the DeepLearning.AI intro course in a few hours, then run a simulated federation the same day. Standing up a real SuperGrid Pro federation with clients, RBAC, and audit logs is a multi-day exercise, longer if you're deploying mobile or C++ clients in the field.
Switching to or from Flower
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a hand-rolled federated setup: move your aggregation logic into a Flower Strategy and keep your existing PyTorch or TensorFlow training code unchanged.
- →From a centralized training pipeline: wrap the dataset partitioning with Flower Datasets and wrap the trainer as a Flower ClientApp, then add a SuperGrid federation.
- →From another federated learning library: port the client update function into Flower's NumPyClient or ClientApp interface; the server strategy is then a drop-in.
- →From a managed no-code AI platform: keep your model code but host the orchestration yourself, since Flower expects you to bring the training loop.
- →From a single-vendor evaluation: the FAB packaging format is stable, so an app you build in simulation runs under SuperGrid without repackaging.
- ↗To a centralized training stack: export the trained model weights and drop the Flower client wrappers — the model artifact is framework-native.
- ↗To a hosted managed AI platform: expect to hand over the orchestration layer, since those platforms typically hide the federation strategy from you.
- ↗To a research federated library: Flower Baselines experiments are configured declaratively, so the strategy definitions translate but the SuperGrid governance layer does not.
- ↗To self-hosted infrastructure: Flower's Kubernetes and Docker support means the same federated app can run on your own cluster instead of SuperGrid.
- ↗To a single cloud AI platform: you would lose framework agnosticism, since most managed platforms lock training to their own stack.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Flower”, and we withheld 6: 6 could not be judged, because “Flower” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Flower.
Official links
Featured Head-to-Head Comparisons
Flower vs Temporal Ai
Choose Temporal AI if you need a battle-tested orchestration engine for AI agents and microservices that must be fault-tolerant and observable. Choose Flower if your priority is federated learning or privacy-preserving collaborative AI on distributed data. Both are complementary: you could use Temporal to orchestrate Flower training rounds across hospitals.
Flower vs Push Security
Push Security is a browser security platform for stopping modern attacks like AiTM phishing and AI data leakage, while Flower is a federated learning framework for privacy-preserving model training. Choose Push if your priority is defending against browser-based threats and controlling AI tool usage; choose Flower if your goal is collaborative AI on sensitive data without centralizing it. Each solves a completely different problem, so the choice depends on whether your need is security or distributed ML.
Flower vs Audioeye
Flower and AudioEye serve completely different needs: Flower is for organizations wanting to train AI on decentralized data without compromising privacy, while AudioEye is for businesses needing to meet web accessibility compliance. Choose Flower if you're building collaborative AI across silos; choose AudioEye if you face ADA/WCAG legal risk.
Popular in Data Labeling & Training Data
Frequently Asked Questions
Used Flower? Help shape our editorial sentiment research.