✨ Features

FlowyML Notebook is a complete ML development environment. Here's everything it offers — with real screenshots from the application.

---

Reactive DAG Engine

Every cell is a node in a dependency graph. When a variable changes, only dependent cells re-execute — no more stale state, no more "restart kernel and run all."

• Automatic dependency detection via AST analysis
• Visual DAG representation with cell status indicators
• Topological execution order for consistent results
• Parallel execution of independent branches

Pipeline DAG: imports → data_generation → analysis → exploration → summary

---

Pure Python Storage

Notebooks are saved as standard .py files. No JSON, no merge conflicts, no diffs you can't read.

• Git Friendly — Clean git diff, meaningful git blame
• Importable — from my_notebook import df_clean
• Lintable — Works with ruff, flake8, mypy

# @cell(id="load_data")
import pandas as pd
df = pd.read_csv("data.csv")

# @cell(id="transform", depends=["load_data"])
df_clean = df.dropna()

---

Rich Data Exploration

Every DataFrame gets automatic profiling — zero extra code. Toggle between Table, Stats, Charts, Correlations, Quality, Insights, Compare, and AI views.

Automatic DataFrame profiling with column statistics, type detection, and memory impact

Interactive charts: histograms, bar charts, and distribution analysis for every column

Pearson correlation matrix with color-coded heatmap

Automated ML insights: outlier detection, scaling recommendations, and target variable suggestions

See Data Exploration for the full breakdown.

---

SmartPrep Advisor — NEW in v1.2

Stop guessing what preprocessing to apply. The SmartPrep Advisor analyzes your DataFrame and generates actionable, ready-to-run suggestions — automatically.

What It Detects

Issue	Detection	Fix
Missing Values	Per-column null counts and percentages	Drop column (>60% missing) or impute (median/mode)
Skewed Distributions	Skewness > 1.5 with positive/negative handling	np.log1p() or PowerTransformer(yeo-johnson)
Outliers	IQR method with percentage thresholds	df.clip(lower, upper)
High Cardinality	>50 unique categorical values	Frequency encoding
Class Imbalance	Majority/minority ratio > 3:1	class_weight='balanced' or SMOTE
Feature Scaling	>100x range difference across features	StandardScaler

Each suggestion includes a severity badge (high/medium/low), a clear explanation, and a "Generate Cell" button that inserts ready-to-run code directly into your notebook.

Usage

Access the SmartPrep tab in the DataFrameExplorer — no code needed. Just click on a DataFrame output and switch to the SmartPrep tab.

# Behind the scenes, the API analyzes your DataFrame:
# GET /api/smartprep/{var_name}?target=y
# Returns severity-ranked suggestions with code snippets

---

Algorithm Matchmaker — NEW in v1.2

Don't know which ML algorithm to use? The Algorithm Matchmaker analyzes your data characteristics, detects the task type, and ranks the best algorithms — with full pipeline code.

How It Works

1. Detects task type — Classification (≤20 unique target values), Regression (>20), or Clustering (no target)
2. Analyzes data — Sample size, feature types, dimensionality, nulls, class balance
3. Ranks algorithms — Score (0-100), speed, interpretability, with reasoning and caveats
4. Generates code — Complete sklearn pipeline: train/test split, fit, evaluate

Supported Algorithms

Task	Algorithms
Classification	Random Forest, XGBoost, Logistic Regression, SVM, LightGBM, KNN
Regression	Random Forest, XGBoost, Linear Regression, SVR, LightGBM, ElasticNet
Clustering	KMeans, DBSCAN, Hierarchical

Usage

Access the Algorithms tab in the DataFrameExplorer. Select a target column and get instant recommendations:

# Example generated pipeline:
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report

X = df.drop(columns=['target'])
y = df['target']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train, y_train)
print(classification_report(y_test, model.predict(X_test)))

---

UnicoLab Keras Ecosystem — NEW in v1.3

Native integration with the UnicoLab ML ecosystem — KDP, KerasFactory, and MLPotion. Install the full stack with:

pip install "flowyml-notebook[keras]"

Ecosystem Integration Points

Package	Integration	What It Adds
KDP	SmartPrep Advisor	Auto-configured Keras preprocessing with distribution-aware encoding and tabular attention
KerasFactory	Algorithm Matchmaker	BaseFeedForwardModel + advanced GatedResidualNetwork / TabularAttention model recommendations
MLPotion	Algorithm Matchmaker	Managed ModelTrainer + ModelTrainingConfig training pipelines
All 3	Algorithm Matchmaker	Flagship end-to-end pipeline: KDP → KerasFactory → MLPotion

Ecosystem Status API

Check installed ecosystem packages and versions:

GET /api/ecosystem/status
→ { packages: [...], installed_count: 3, fully_integrated: true, install_all: "pip install 'flowyml-notebook[keras]'" }

Builtin Ecosystem Recipes

4 new multi-cell recipes available in the recipe library — KDP Smart Preprocessing, KerasFactory Quick Model, MLPotion Training Pipeline, and the UnicoLab End-to-End Pipeline.

See Ecosystem for the full guide with code examples.

---

Live Interactive Dashboards — NEW in v1.2

Transform any notebook into a stakeholder-ready interactive dashboard — with auto-detected widgets, auto-refresh, shareable URLs, and email snapshots.

Dashboard Features

Feature	Description
Interactive Widgets	Auto-detects DataFrame columns and creates filters, sliders, and dropdowns
Auto-Refresh	Configurable interval (30s, 1min, 5min, 15min) for live data
Shareable URLs	Generate links with embedded dashboard state (filters, selections)
Email Snapshots	Send a screenshot of the current dashboard state via email

Widget Types

• Date Range — For datetime columns
• Category Dropdown — For categorical columns
• Numeric Slider — For numeric columns
• Text Search — For text/string columns

Access via the App Publisher panel → enable "Interactive Widgets" toggle.

---

Collaborative Analysis Patterns — NEW in v1.2

Bookmark reusable cell sequences as Analysis Patterns — share them across notebooks, search by tags, and apply with one click.

Pattern Features

• Create — Select cells from your notebook, add tags and metadata, save as a pattern
• Search — Filter by name, tags, problem type (classification, regression, clustering, EDA), or data type
• Apply — One-click inserts all pattern cells into your notebook
• Usage tracking — See how many times each pattern has been applied

Creating a Pattern

1. Open the Analysis Patterns panel (right panel)
2. Click New Pattern
3. Select which cells to include
4. Add name, description, tags, problem type, and data type
5. Click Save Pattern

Example Patterns

Pattern	Tags	Problem Type
EDA Starter	eda, pandas, visualization	EDA
Feature Engineering Pipeline	features, sklearn, preprocessing	Any
Binary Classification Baseline	classification, xgboost, metrics	Classification
Time Series Analysis	time_series, decomposition, stationarity	Regression

API

# List patterns
GET /api/patterns

# Save new pattern
POST /api/patterns  {name, description, tags, cells, problem_type, data_type}

# Apply pattern
POST /api/patterns/{id}/apply

# Search patterns
POST /api/patterns/search  {query, tags, problem_type, data_type}

---

AI Assistant

Integrated AI assistant (⌘J) with deep FlowyML ecosystem knowledge. The assistant is context-aware — it sees your notebook state, cell outputs, variable values, and error tracebacks.

• Generate — Create pipeline segments, data transformations, or visualizations from natural language
• Explain — Understand complex data patterns from outputs and profiling results
• Debug — Context-aware error resolution with stack trace analysis
• Optimize — Performance tuning, scaling recommendations, and ML hyperparameter suggestions

Supported Providers

Provider	Models	Setup
OpenAI	GPT-4o, GPT-4o-mini	OPENAI_API_KEY env variable
Google AI	Gemini Pro, Gemini Ultra	GOOGLE_API_KEY env variable
Ollama	Llama 3.1, Mistral, CodeLlama, Phi-3	Local — ollama serve (no API key needed)
Anthropic	Claude 3.5 Sonnet, Claude 3 Opus	ANTHROPIC_API_KEY env variable
Custom	Any OpenAI-compatible API	Set base_url to your endpoint

100% Private with Ollama

Run AI entirely on your machine — no data ever leaves your laptop:

# Install Ollama: https://ollama.com
ollama pull llama3.1
ollama serve
# FlowyML auto-detects Ollama at localhost:11434

Usage Examples

# In the AI chat panel:
> Load a CSV, clean missing values, and plot the distribution of 'revenue'
> Why did cell 3 fail with a KeyError?
> Optimize this XGBoost model for better F1 score
> Write a FlowyML pipeline step that processes this DataFrame

The AI assistant generates code that respects the reactive DAG — it understands which variables are available from upstream cells and avoids creating stale dependencies.

---

Comments & Review

Collaborate directly in the notebook with inline comments. Add notebook-level or cell-level annotations for team discussions, with resolve/reply threading.

Comments panel with threaded discussions alongside code and outputs

---

Recipes — Reusable Templates

43 built-in recipes across Core, Assets, Parallel, Observability, Evals, Data, ML, Visualization, and Ecosystem categories. Drag into your notebook or click + to insert.

Searchable recipe library with FlowyML Step, Pipeline, Conditional Branching, and more

See Recipes for the full cookbook.

---

Reports

Generate beautiful dark-themed HTML reports from your notebook with one click. Reports include styled tables with inline SVG histograms, correlation badges, and formatted outputs.

One-click report generation with HTML/PDF format, code inclusion, and browser preview

Report Features

• Dark-themed — Premium dark UI with Inter + JetBrains Mono fonts
• DataFrame rendering — Auto-generated stat cards, SVG histograms, and categorical bar charts inline
• Code toggle — Optionally include source code alongside outputs
• Markdown cells — Rendered as formatted headings, lists, and paragraphs
• CLI export — fml-notebook export notebook.py --format html --output report.html

Via Python API

from flowyml_notebook.reporting import generate_report

generate_report(
    notebook=nb,
    format="html",         # or "pdf"
    include_code=False,    # Show outputs only
    title="Monthly Fraud Analysis",
    output_path="report.html",
)

---

Publish as App

Turn any notebook into an interactive web application with one click. Choose your layout, toggle cell visibility, and deploy:

App publishing with layout options, dark/light theme, and per-cell visibility control

Layout Options

Layout	Description
Linear	Cells stacked vertically — classic report layout
Grid	Responsive grid — cells arranged in columns
Tabs	Each cell is a tab — ideal for dashboards
Sidebar	Input controls in sidebar, outputs in main area
Dashboard	Full dashboard mode with widgets and charts

Via CLI

fml-notebook app my_notebook.py --layout dashboard --port 8501

Widgets (sliders, dropdowns, toggles) become interactive controls in the deployed app — users can tweak parameters and see results update in real time.

---

Production — Pipelines, Deploy & Assets

Ship notebooks directly to production:

PipelinesDeployAssets

Promote notebooks to production FlowyML pipelines with @step decorators.

Pipeline promotion with quick actions: Export, Run All, View DAG, HTML Report

Deploy as REST API, Docker Container, or Batch Pipeline — with full FlowyML infrastructure stack integration.

Production deployment: API (recommended), Docker, Batch Pipeline, plus infrastructure stacks

Track kernel assets (DataFrames, models) with size, shape, and type metadata.

Kernel assets: tracked DataFrames with size, rows × cols, and type tags

---

Git & Version Control

Full GitHub integration as the collaboration backend. No proprietary cloud, no database — just Git.

Connect GitHub repository for team collaboration, versioning, and shared recipes

Save and browse notebook snapshots with cell-level diffs

See Collaboration for the full workflow.

---

Environment & Connection

Run standalone in Local Mode or connect to a remote FlowyML server for experiment tracking, pipeline export, and deployment.

Local/Remote connection, Python 3.12 runtime, IPython kernel, Reactive DAG engine

See Integration for connection details.

---

SQL First-Class

Mix Python and SQL seamlessly in the same notebook. SQL cells are powered by DuckDB (in-memory, zero config) or SQLAlchemy (for external databases).

-- @cell(type="sql")
SELECT model, AVG(accuracy) as avg_acc, COUNT(*) as n
FROM df_experiments
WHERE accuracy > 0.85
GROUP BY model
ORDER BY avg_acc DESC

How It Works

• SQL cells query your Python DataFrames directly — any variable holding a DataFrame is available as a table name
• Results are automatically converted to Pandas/Polars DataFrames
• The result DataFrame is assigned to a variable and participates in the reactive DAG
• Supports DuckDB (default, in-process) and SQLAlchemy (PostgreSQL, MySQL, SQLite, etc.)

External Databases

-- @cell(type="sql", connection="postgresql://user:pass@host/db")
SELECT * FROM users WHERE created_at > '2026-01-01'

---

Interactive Widgets

Bind Python variables to professional-grade UI controls — sliders, dropdowns, toggles, date pickers, color pickers, text inputs, and more.

import flowyml_notebook.ui as fml

# Slider with range and step
learning_rate = fml.slider("Learning Rate", min=0.0001, max=0.1, step=0.0001, default=0.001)

# Dropdown selection
model_type = fml.select("Model", options=["XGBoost", "LightGBM", "SVM", "Neural Net"])

# Toggle switch
use_augmentation = fml.toggle("Data Augmentation", default=True)

# Number input
n_estimators = fml.number("Trees", min=10, max=1000, default=100)

Widget Reactivity

Widgets participate in the reactive DAG — when a user changes a slider value, all downstream cells automatically re-execute. This is what powers the Publish as App feature: your notebook becomes a live, interactive dashboard.

Available Widgets

Widget	Description	Code
Slider	Numeric range	fml.slider(...)
Select	Dropdown menu	fml.select(...)
Toggle	Boolean switch	fml.toggle(...)
Number	Numeric input	fml.number(...)
Text	Text input	fml.text(...)
Date	Date picker	fml.date_picker(...)
Color	Color picker	fml.color(...)
File	File upload	fml.file_upload(...)

---

Deployment & Export

Format	Command	Use Case
FlowyML Pipeline	fml-notebook export --format pipeline	Production ML pipelines
HTML Report	fml-notebook export --format html	Shareable dashboards
PDF Report	fml-notebook export --format pdf	Documentation
Docker Image	fml-notebook export --format docker	Containerized deployment
Web App	fml-notebook app notebook.py	Interactive applications