NEExT Workbench

The Workbench is an optional local, single-user desktop-style web app over the NEExT library — a FastAPI backend serving a React UI. It’s for running and inspecting NEExT workflows interactively without writing a script for every step. The core pip install NEExT stays lightweight; the Workbench lives behind an extra.

Install & launch

shell

pip install "NEExT[workbench]"
neext-workbench

This starts the server and opens your browser at http://127.0.0.1:8765.

CLI options

Flag	Default	Meaning
`--workspace`	`~/NEExT-Workbench`	Folder for Workbench projects (override with `NEEXT_WORKBENCH_HOME`)
`--host`	`127.0.0.1`	Bind host — keep localhost for trusted local execution
`--port`	`8765`	Preferred port
`--no-browser`	off	Don’t open a browser window on launch

shell

neext-workbench --workspace ~/research/wb --port 9000 --no-browser

The interface

The Workbench is organized into Spaces along the top — Home, Datasets, Features, Embeddings, and Models. Each Space has a Ribbon of grouped commands, and the main work happens in Center Views in the center panel:

Left Panel — your project’s artifacts and the current selection.
Center Panel — the active Center View (libraries, forms, previews, analysis).
Right Panel — system info: the Inspector (selected-item details) and the Jobs queue.
Command Window — logs, errors, and job output along the bottom.

The workflow

The Workbench mirrors the library’s pipeline as a chain of saved artifacts, each one the input to the next:

Project — the top-level container.
Dataset — imported graph tables (the NEExT edge / node-mapping contract), prepared into NEExT-ready graphs.
Feature — structural node features (built-in or custom Python) computed on a dataset.
Embedding — graph embeddings computed from one or more features.
Model — a supervised model trained on an embedding, with metrics and feature-importance analysis.

Running an artifact creates a job; the single local worker executes it and streams progress to the Jobs panel and Command Window. Datasets, features, embeddings, and models also have analysis views (clustering, projections, feature importance).

Workspace on disk

Everything is stored as plain files under the workspace folder, so projects are portable and inspectable:

~/NEExT-Workbench/

workspace.json
projects/
<project_uuid>/
  project.json
  artifacts/
    datasets/   <dataset_uuid>/
    features/   <feature_uuid>/
    embeddings/ <embedding_uuid>/
    models/     <model_uuid>/
  jobs/
    <job_uuid>/
trash/
projects/
  <project_uuid>/

Want to drive the Workbench from an AI agent? See Workbench MCP.