Drag intelligent nodes onto a canvas, connect them with typed edges, and watch ConduitCraft AI generate executable Python, Jupyter notebooks, Kubeflow DSL, and Dockerfiles — no boilerplate, no guesswork.
Watch an end-to-end ML pipeline in action — ingest a CSV dataset, scale features, train a Random Forest classifier, evaluate metrics, and register the model to MLflow.
Every feature is purpose-built for the ML and LLM development workflow — from first prototype to production deployment.
Drag nodes from a categorised palette onto a React Flow canvas. Connect them with colour-coded typed edges — incompatible port types are blocked at draw time. No runtime surprises, no boilerplate setup.
One click produces Python scripts, Jupyter notebooks, Kubeflow v2 DSL, or Dockerfiles with all imports and pip packages resolved and listed automatically.
After a run each node shows its output directly on the canvas card — DataFrame shape + mini table, model class name, metrics, or text preview. No Jupyter context-switch.
Built-in ingest nodes for local files, AWS S3, Azure Blob, GCS, PostgreSQL, and HuggingFace Datasets — with Test Connection and live row preview in the Inspector.
Run locally with your Python environment or inside Docker. Real-time WebSocket log streaming with per-node status badges — running, success, or failed — visible on the canvas.
Every edge is typed — DataFrame, Model, Embeddings, VectorStore, Text, Metrics, or Any. Incompatible handles dim automatically. The Validate button scans the whole pipeline in one click.
Community plugins add new node types and connectors via a manifest-first, iframe sandbox model — communicate with the host using a postMessage API, no core changes needed.
An Electron wrapper bundles the FastAPI backend as a self-contained PyInstaller executable for Windows, macOS, and Linux. No Python install required on the end-user machine — just download and run.
Two purpose-built pipeline modes with dedicated node palettes, each covering the full lifecycle from raw data to deployed model.
CSV, Parquet, S3, Azure Blob, GCS, PostgreSQL, HuggingFace Datasets
Standard Scaler, Encoder, Column Filter, Outlier Removal, Train/Test Split
Random Forest, XGBoost, Gradient Boosting, Logistic Regression, Keras CNN, PyTorch Trainer
Classification & Regression Metrics · MLflow Registry · HuggingFace Hub · Docker · Evidently Drift
PDF, Web Pages, S3 Docs · Recursive Text Splitter · OpenAI / HuggingFace / Ollama Embeddings
Chroma, FAISS, Pinecone, Weaviate, pgvector
OpenAI GPT-4o · Anthropic Claude · Ollama (local) · vLLM · HuggingFace Inference
RAG Chain · ReAct Agent · LangGraph Workflow · LlamaIndex Query · LangServe · FastAPI
The pipeline is the single source of truth. Topological sort → Jinja2 template render per node → assembled output in your chosen format.
# Auto-generated by ConduitCraft AI
# pip install pandas scikit-learn mlflow
import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
import mlflow
def ingest_csv():
return pd.read_csv("data/iris.csv")
def transform_scaler(df):
scaler = StandardScaler()
X = scaler.fit_transform(df.drop("species", axis=1))
return X, df["species"]
def train_random_forest(X, y):
with mlflow.start_run():
model = RandomForestClassifier(n_estimators=100)
model.fit(X, y)
mlflow.sklearn.log_model(model, "model")
return model
if __name__ == "__main__":
df = ingest_csv()
X, y = transform_scaler(df)
model = train_random_forest(X, y)Native support for the tools and platforms you already use — configured once in the Integrations panel, applied everywhere.
Native desktop app for Windows, macOS, and Linux — no Python install required. Bundled FastAPI backend included.
Open source. Apache 2.0. Runs entirely on your machine.