Flower
Hamburg federated learning framework with a hosted SuperGrid service, for training across sites without centralising data
Quick Overview
| Company | Flower Labs GmbH |
|---|---|
| Category | MLOps Platforms |
| Headquarters | Hamburg, Germany |
| Founded | Not stated |
| EU Presence | Yes - Germany (EU) |
| Data Location | Training stays on each participant's own data; SuperGrid control plane is hosted by Flower Labs |
| Open Source | Apache-2.0 (flower repository on GitHub) |
| GDPR Compliant | Yes |
| Self-Hosting | Yes - open source framework |
| Main Features | Federated learning framework, SuperGrid federations, web UI, public and private federations, audit logs, RBAC, Flower Hub, confidential compute on Max |
| Pricing | SuperGrid paid plans in euros per month billed yearly with monthly credits; 3,000 sign-up credits; Max on request |
| Best For | Organisations training across hospitals, banks or devices without sharing raw data |
| Replaces | Centralised training pipelines in Weights & Biases, SageMaker |
Detailed Review
Flower tackles a problem general MLOps tools skip: training a model where the data cannot be copied to one place. Each participant trains locally and only model updates are combined.
The imprint names Flower Labs GmbH, Winterhuder Weg 29, Hamburg, registered at Amtsgericht Hamburg HRB 160919. The Terms of Service for the SuperGrid SaaS were last edited on 13 April 2026.
Framework and hosted grid
The framework is Apache-2.0 on GitHub. SuperGrid is the hosted service that runs federations, with a web UI, public and private federations, audit logs and RBAC.
Higher tiers add advanced RBAC, verified apps, confidential compute and private apps on Flower Hub.
Credit-based pricing
SuperGrid plans are priced in euros per month, billed yearly, with monthly credit allowances, and new accounts can claim 3,000 sign-up credits. The Max plan is custom.
Because usage is metered in credits, estimating cost for a long training job needs a test run.
Direction of the company
The home page now describes Flower Labs as a full-stack NeoLab, with Flower Agent and Flower Models next to SuperGrid. The Terms say the cloud service supports LLM inference and, in future, tuning and training.
That roadmap is relevant if you are buying for the long term.
Limitations to Consider
Flower is not a general MLOps platform: there is no experiment dashboard, registry or orchestration layer, so it is a specialist piece beside ZenML or DagsHub. Credit-based pricing makes cost estimates harder. The company is broadening into agents and models, so the hosted service may change. Federated projects also need participants who agree on infrastructure and governance, which is organisational work no tool removes.
Who Should Use Flower
Healthcare, finance and device-fleet teams in Europe that need to train on data that must stay where it is.
Pros and Cons
Pros
- Apache-2.0 framework with a large community
- Training without centralising raw data
- German company with Hamburg register number
- Euro prices on the hosted service
- Audit logs and RBAC in SuperGrid
Cons
- Specialist tool, not a full MLOps platform
- Credit-based pricing is hard to forecast
- Company direction is widening
- Federations need multi-party coordination
Alternatives to Flower
Other European mlops platforms worth comparing before you decide:
ZenML
Germany · Munich open source pipeline orchestration with artifact and model tracking
Valohai
Finland · Turku platform for versioned ML runs, pipelines and GPU scheduling
Hopsworks
Sweden · Stockholm feature store with model registry and serving, free tier
DagsHub
Israel · Experiment tracking, data versioning and annotation on one platform
See all European mlops platforms or compare them directly against Centralised training pipelines in Weights & Biases on our alternatives page.
Frequently Asked Questions
It trains a shared model across several sites where each keeps its data locally and only model updates are combined.
Yes. The framework is published under Apache-2.0 on GitHub, and SuperGrid is the hosted commercial service.
No. It handles distributed training across sites, so you still need a tracking and registry tool for the results.