Connectify Docs
Login Start building
Feature

Run Experiments

Execute graphs, capture results, and compare across variants. Run Experiments is where Connectify earns its keep as an ablation tool.

Variants

A variant is an independent copy of your graph that shares the project but has its own structure, config, and history. Use variants to:

Creating a variant

Click the + Variant tab in the variant strip at the top of the editor. The new variant starts as a clone of whichever variant you were viewing. Rename it to something meaningful — "Baseline", "Higher LR", "No augmentation".

Switching variants

Click any variant tab. Connectify saves your in-flight changes, swaps the graph, history, and paths, and reopens the new variant in the same state you left it.

Baselines

Designate one variant as the baseline (right-click the tab → Set as baseline). Other variants get a "vs. Baseline" comparison option in their run results.

Running a graph

Click Run in the top-right toolbar. Connectify executes the graph in topological order:

  1. Each node's inputs resolve from upstream outputs.
  2. The node runs its computation.
  3. Outputs propagate to the next layer.

Edges animate as data flows through them — green pulse while data is in transit, solid green when the downstream node finishes.

In-browser execution

Lightweight graphs run entirely client-side. Heavier graphs (large models, big datasets) offload to a runtime worker. The run button shows a progress indicator and is cancelable.

Reading run data

After a run, click any node and open the Run Data tab in the Inspector. You'll see:

Comparing variants

From any variant, open the Compare dropdown in the toolbar. Pick a second variant to diff against. You'll see:

Compare highlights the differing nodes in both canvases so you can scan visually.

Hiding variants

As your variant strip grows, hide the ones you're not actively viewing. Right-click a tab → Hide. Hidden variants don't appear in the strip but stay in the project; toggle visibility from Variants → Manage.