🌐 The FlowyML Ecosystem
FlowyML-Universe provides a complete, end-to-end suite of tools for the modern ML team — from interactive experimentation in a reactive notebook, through artifact-centric pipeline orchestration, to one-click cloud deployment. Every tool is designed to work together seamlessly so you can focus on the science, not the plumbing.
📓 FlowyML Notebook NEW
The Reactive Notebook That Ships to Production
FlowyML Notebook is a reactive, DAG-powered notebook that replaces Jupyter for production ML workflows. Write pure Python cells, get automatic dependency tracking, and ship directly to FlowyML pipelines, dashboards, and interactive apps — all without leaving the notebook.
🚀 Why FlowyML Notebook?
Jupyter gave us exploration. Marimo gave us reactivity. FlowyML Notebook gives you both — plus a direct bridge to production pipelines, one-click deployments, and an AI-powered development experience that no other notebook offers.
⚡ Quick Start
🧬 Core Capabilities
⚡ Reactive DAG Engine
Cells form a dependency graph. Change a variable, and only the dependent cells re-execute — no stale state, no manual re-runs. Think spreadsheet-meets-notebook.
🐍 Pure .py Storage
Notebooks are saved as plain Python files — not .ipynb JSON. That means real git diff, linting, importing, and code review. Your notebooks are now first-class code.
📊 Rich Data Exploration
Every DataFrame gets an automatic 10-tab profiling panel — statistics, distributions, correlations, missing values, and more. Understand your data before you model it.
🧹 SmartPrep Advisor PRO
Auto-detects data quality issues — missing values, outliers, skew, cardinality — and generates one-click fix code directly into your notebook. Data cleaning on autopilot.
🎯 Algorithm Matchmaker PRO
Tell it your task type and target column. It auto-ranks ML algorithms, benchmarks them, and generates ready-to-run scikit-learn pipeline code tailored to your data.
📚 43 Built-in Recipes
Reusable code templates across 9 categories — data loading, preprocessing, modeling, evaluation, visualization, deployment, and more. Stop writing boilerplate.
🔧 Full Feature Set
🔗 GitHub Integration
Branch, commit, push, and pull from the notebook sidebar. No terminal needed — version control is built into your workflow.
🚀 Pipeline Promotion
Promote notebooks directly to production FlowyML pipelines with a single click. Your experiment becomes your deployment.
📦 One-Click Deploy
Export as REST API, Docker container, HTML/PDF report, or interactive web app. From notebook to production in seconds.
🤖 AI Assistant
Context-aware code generation powered by OpenAI, Google AI, Ollama, or Anthropic. It knows your data, your cells, and your intent.
🗄️ SQL First-Class
Mixed Python + SQL cells with built-in DuckDB for in-memory analytics or SQLAlchemy for remote databases. Query results flow into the DAG.
🖥️ App Mode
Turn any notebook into a web application with 5 layout options — dashboard, report, slides, sidebar, and custom. Share insights, not code.
🧠 Keras Ecosystem
Native integration with KDP, KerasFactory, and MLPotion — the UnicoLab Keras stack is a first-class citizen inside the notebook.
🔄 Live Collaboration BETA
Real-time multiplayer editing with cursor presence, cell locking, and conflict resolution. Pair-program inside the notebook.
📈 Experiment Tracking
Every run is automatically logged — parameters, metrics, artifacts, and lineage. Compare experiments without leaving the notebook.
📊 FlowyML Notebook vs. The Competition
| Feature | Jupyter | Deepnote | Marimo | FlowyML Notebook |
|---|---|---|---|---|
| Reactive DAG | ❌ | ❌ | ✅ | ✅ |
| Pure .py Storage | ❌ | ❌ | ✅ | ✅ |
| Git-Native | ❌ | ⚠️ | ❌ | ✅ |
| Pipeline Integration | ❌ | ❌ | ❌ | ✅ |
| Reusable Recipes | ❌ | ❌ | ❌ | ✅ |
| One-Click Deploy | ❌ | ⚠️ | ❌ | ✅ |
| SQL First-Class | ❌ | ✅ | ✅ | ✅ |
| AI Assistant | ❌ | ✅ | ❌ | ✅ |
| Self-Hosted | ✅ | ❌ | ✅ | ✅ |
| SmartPrep Advisor | ❌ | ❌ | ❌ | ✅ |
| Algorithm Matchmaker | ❌ | ❌ | ❌ | ✅ |
FlowyML Notebook is the only notebook that combines reactivity, pure Python storage, and direct pipeline integration with production ML orchestration.
🔗 Links & Resources
🧠 UnicoLab Keras Ecosystem
The UnicoLab Keras Ecosystem is a trio of libraries that supercharges Keras and TensorFlow workflows — from data preprocessing to model architecture to managed training. Every component integrates natively with FlowyML Notebook and FlowyML Pipelines.
🔮 KDP — Keras Data Processor
Smart preprocessing layers that plug directly into your Keras model. Handle tabular features, text, images, and mixed-modality inputs with learnable preprocessing — no separate pipeline needed.
from kdp import PreprocessingModel
preprocessor = PreprocessingModel(
path_data="train.csv",
features_specs={
"age": FeatureType.FLOAT_NORMALIZED,
"city": FeatureType.STRING_CATEGORICAL,
"bio": FeatureType.TEXT,
}
)
model = preprocessor.build_preprocessor()
PRO Learnable preprocessing inside the model graph
🏭 KerasFactory
38+ reusable Keras layers and architectures — attention blocks, residual connections, transformer encoders, tabular heads, and more. Stop reimplementing the same layers project after project.
from keras_factory import TabularModel
model = TabularModel(
numerical_features=["age", "income"],
categorical_features=["city", "plan"],
architecture="residual_attention",
output_units=1,
output_activation="sigmoid",
)
model.compile(optimizer="adam", loss="binary_crossentropy")
NEW 38+ production-tested layer recipes
🧪 MLPotion
Managed training pipelines for Keras models — experiment tracking, hyperparameter sweeps, early stopping strategies, and reproducible training runs, all configurable via YAML or Python.
from mlpotion import TrainingPipeline
pipeline = TrainingPipeline(
model=model,
data=dataset,
config="training_config.yaml",
experiment_name="fraud_detection_v3",
)
results = pipeline.run()
print(results.best_metrics)
BETA YAML-driven training orchestration
🔗 End-to-End Keras Workflow
KDP prepares your data → KerasFactory builds your architecture → MLPotion manages training → FlowyML orchestrates everything in production. All from a single FlowyML Notebook.
🔌 Integration Ecosystem
FlowyML connects to the tools you already use. Swap backends, cloud providers, and experiment trackers without touching your pipeline code.
🤖 ML Frameworks
First-class support for the leading ML frameworks — use any of them inside FlowyML steps.
- Keras / TensorFlow — Native callbacks & asset types
- PyTorch — Checkpoint tracking & model serialization
- Scikit-learn — Pipeline wrapping & artifact capture
- HuggingFace 🤗 — Tokenizers, models, and datasets
- XGBoost / LightGBM — Tree-based model support
📊 Experiment Tracking
Log metrics, parameters, and artifacts to your tracking platform of choice.
- MLflow — Full experiment tracker plugin
- Weights & Biases — Native W&B integration
- FlowyML UI — Built-in dashboard at
localhost:8080 - TensorBoard — Keras callback support
☁️ Cloud Providers
Same pipeline code, different infrastructure. One YAML change to switch clouds.
- GCP — Vertex AI orchestration, GCS artifact store, Vertex Model Registry
- AWS — SageMaker orchestration, S3 artifact store
- Azure — Azure ML orchestration, Blob Storage artifact store
🐳 Infrastructure
Docker containers, Kubernetes orchestration, and Helm chart deployment — run anywhere at any scale.
💬 Communication
Slack notifications, email alerts, and webhook integrations — keep your team informed on every pipeline event.
🗃️ Data & Storage
DuckDB, PostgreSQL, BigQuery, Snowflake, and file-based stores — connect to any data source your pipelines need.
🔄 How They Work Together
The FlowyML ecosystem is designed as a continuous loop — from experimentation to production and back. Here's how the pieces connect:
graph TD
NB["📓 FlowyML Notebook<br/><i>Design & Experiment</i>"]
KDP["🔮 KDP<br/><i>Smart Preprocessing</i>"]
KF["🏭 KerasFactory<br/><i>Model Architecture</i>"]
MLP["🧪 MLPotion<br/><i>Managed Training</i>"]
NB -->|"Build models with"| KDP
NB -->|"Design architectures with"| KF
NB -->|"Train models with"| MLP
NB ==>|"🚀 Pipeline Promotion"| FM
FM["🌊 FlowyML<br/><i>Orchestrate & Deploy</i>"]
FM -->|"Stacks & Plugins"| CLOUD
subgraph CLOUD["☁️ Infrastructure Layer"]
GCP["GCP<br/>Vertex AI"]
AWS["AWS<br/>SageMaker"]
AZ["Azure<br/>Azure ML"]
MLF["MLflow<br/>Tracking"]
K8S["Kubernetes<br/>Orchestration"]
end
CLOUD -->|"Monitoring & Feedback"| NB
style NB fill:#6366f1,stroke:#4f46e5,color:#fff
style FM fill:#8b5cf6,stroke:#7c3aed,color:#fff
style KDP fill:#10b981,stroke:#059669,color:#fff
style KF fill:#10b981,stroke:#059669,color:#fff
style MLP fill:#10b981,stroke:#059669,color:#fff
style CLOUD fill:#1e1b4b,stroke:#312e81,color:#c7d2fe
style GCP fill:#1e293b,stroke:#334155,color:#e2e8f0
style AWS fill:#1e293b,stroke:#334155,color:#e2e8f0
style AZ fill:#1e293b,stroke:#334155,color:#e2e8f0
style MLF fill:#1e293b,stroke:#334155,color:#e2e8f0
style K8S fill:#1e293b,stroke:#334155,color:#e2e8f0🔁 The Development Loop
1. Explore & Design
Open FlowyML Notebook, load data, run SmartPrep Advisor, and prototype your model using KDP + KerasFactory layers. The reactive DAG keeps every cell fresh.
2. Train & Evaluate
Use MLPotion or the built-in Algorithm Matchmaker to run training sweeps. Compare results with the 10-tab profiler and experiment tracking.
3. Promote to Pipeline
One click to promote your notebook into a FlowyML production pipeline. Steps, artifacts, and dependencies are auto-extracted from your cells.
4. Orchestrate & Deploy
FlowyML runs your pipeline on Vertex AI, SageMaker, or Kubernetes. Artifacts are routed to GCS, S3, or MLflow based on your stack config.
5. Monitor & Iterate
Track data drift, model performance, and LLM traces in the FlowyML dashboard. When things change — loop back to the notebook.
🏁 Get Started
📓 Start with the Notebook
The fastest path to production ML.
🌊 Start with Pipelines
Already have code? Orchestrate it.
🧠 Explore the Keras Stack
Supercharge your Keras workflows.
One ecosystem. Every stage of ML.
Design in the notebook. Ship with the pipeline. Scale on the cloud.