Skip to content

Skyulf

The Visual MLOps Builder — Drag-and-drop machine learning without writing code.

Skyulf is a self-hosted, privacy-first MLOps platform. Bring your data, clean it visually, engineer features with a node-based canvas, train models, and deploy — all in one place.

Why Skyulf?

Feature What you get
Visual ML Canvas Build end-to-end pipelines by connecting nodes on a React Flow canvas
20+ Models Classification & Regression (Random Forest, XGBoost, SVM, KNN, and more)
Advanced Tuning Grid, Random, Successive Halving, and Optuna with configurable Samplers & Pruners
Automated EDA Profiling, outlier detection, PCA, causal discovery, drift monitoring
Hybrid Engine Polars for fast ingestion, Pandas/Scikit-Learn for ML compatibility
Leakage Prevention Calculator/Applier architecture ensures train-only statistics
One-Click Deploy Serve models via REST API with built-in inference testing

Quick Install

skyulf-core (standalone Python library)

pip install skyulf-core

Full platform (backend + frontend)

git clone https://github.com/flyingriverhorse/Skyulf.git
cd Skyulf
pip install -r requirements-fastapi.txt
python run_skyulf.py

Open http://127.0.0.1:8000 to access the dashboard.

Architecture

Skyulf is built on three components with strict boundaries:

  • Frontend: React + TypeScript + React Flow (visual ML canvas)
  • Backend: FastAPI + Celery + Redis (API, jobs, persistence)
  • Core Library: skyulf-core (standalone Python package on PyPI)
  • Data Engine: Hybrid Polars (high-performance ingestion) + Pandas (ML ecosystem)
frontend  →  (HTTP)  →  backend  →  (import)  →  skyulf-core

See Architecture and Data Architecture for deep dives.

Where to Start

Using the Web Platform

  1. Follow the Platform Setup to get the backend running (Docker or manual).
  2. See the Platform Walkthrough for an 8-step end-to-end guide through the UI.
  3. Explore the interactive API docs at /docs (Swagger UI) or /redoc (ReDoc).

Using skyulf-core as a Library

  1. Installation — Install from PyPI or editable mode.
  2. Overview — Understand the Calculator/Applier pattern.
  3. Pipeline Quickstart — Build your first pipeline in Python.
  4. Step-by-Step (No Config) — Low-level building blocks.
  5. Reference → Preprocessing Nodes / Modeling Nodes — Full node catalog.

Going Deeper