Glossary

Adaptor

A small named conversion that sits on an edge between two ports with different types. Bridges, for example, jpg → png or tensor → image. See Type System & Adaptors.

Baseline

A variant designated as the reference point for comparisons. Other variants can diff their results against the baseline.

Canvas

The infinite 2D surface where nodes and edges live. Pan with space-drag, zoom with Cmd/Ctrl-scroll.

Checkpoint

A snapshot of a trained model's weights. Produced by Model nodes after a training run; usable as the starting point for further training or as the deployable artifact.

Comment

A threaded message attached to a specific node. Visible in the Inspector's Comments tab; useful for review and async discussion.

Connection

See Edge. The two terms are interchangeable in Connectify, though "edge" is the visual element and "connection" is the underlying data record.

Custom node

A user-authored node implemented as a Python plugin. See Extend with Plugins.

Dataset node

A node type that produces data. Always a root in the graph — has no inputs.

Edge

A wire connecting an output port on one node to an input port on another. May carry an adaptor.

Fork

An independent copy of a graph in your own workspace. Forking from a public graph or a teammate's project gives you a sandbox that doesn't affect the original.

Graph

The whole computational pipeline: nodes, edges, subgraphs, and variants together. A graph belongs to a project.

Inspector

The right-side panel that shows a node's configuration and run data. Opens when you click a node; closes on Escape or canvas click.

Logic node

A node type for deterministic transformations: filters, joins, normalizers, metrics. Everything that isn't a Dataset, Model, or Custom node.

Marquee

A selection rectangle drawn by dragging on empty canvas. Every node inside the rectangle joins the selection.

Model node

A node type wrapping a trained or trainable model. Can run in inference or training mode.

Node

A computational unit on the canvas. Every node has a type (Dataset, Model, Logic, Custom), input ports, output ports, and a config block.

Path

A saved trace through the graph from an output back to the inputs that feed it. Used for debugging, documentation, and review. See Path Tracing.

Plugin

A Python module that registers Custom nodes, types, or adaptors. Lives in the workspace's plugins/ directory.

Port

A typed input or output on a node. Inputs sit on the left edge of a node, outputs on the right. Connect two ports by dragging from an output to an input.

Project

The container for a graph and all its variants, runs, and history. Sits in your workspace and has its own sharing permissions.

Run

A single execution of a graph. Produces run data on every node, animated edges, and a metric in any terminal output.

Run data

The inputs, outputs, and intermediate variables a node observed during a run. Visible in the Inspector's Run Data tab.

Slug

A stable, URL-safe identifier for a node. Survives label renames so connections don't break when you rename things.

Subgraph

A group of nodes wrapped into a single collapsible tile. Useful for keeping a dense canvas readable.

Tag

A free-form label attached to a node, variant, or project. Used for filtering and organization.

Type

The semantic kind of data a port carries — image, tensor, float, string, label, etc. Two ports must have compatible types (or an adaptor between them) to connect.

Variant

An independent copy of a graph within the same project. Each variant has its own structure, config, history, and runs. Use variants for ablations and sweeps.

View Mode

A read-only rendering of a graph. The same canvas, the same paths, the same run data — but no edits. Used for sharing with stakeholders and previewing public graphs before forking.

Workspace

The top-level container for projects, plugins, members, and settings. Personal accounts have one workspace; teams have a shared workspace with role-based permissions.