Skip to main content
When you research a complex topic, a single AI thread quickly becomes a wall of text where labor economics, healthcare data, and education outcomes all blur together. NodePad lets you split a question into focused sub-investigations and run them side by side — each thread sharp on its own topic, all of them visible at once.

​
Why the canvas changes how you research

A typical AI chat collapses everything into one chronological scroll. When you ask follow-up questions, the model inherits the whole conversation, including tangents that dilute focus. On NodePad’s canvas, each thread is spatially distinct. You see them all at once, compare findings across neighbors, and keep each model anchored to exactly one sub-question.

Parallel threads

Run separate investigations simultaneously instead of asking one model to juggle everything in a single chat.

Right model per question

Pick Claude Opus for deep analysis, Gemini 2.5 Pro for data retrieval, GPT-5 for broader synthesis — per thread.

Merge when ready

Pull findings from multiple threads into one synthesis node you write yourself, preserving full context.

​
Walkthrough: UBI policy research

Here is how a three-thread UBI investigation looks on a single canvas.
1

Open a new canvas and start your main framing thread

Create a canvas for your research project. In the first node, ask your framing question: what are the key dimensions of a universal basic income policy analysis? Use this thread to scope the problem and identify the sub-questions you want to investigate separately.
Attach a sticky note to this node with the overall scope constraint — for example, “focus on peer-reviewed studies and government pilot programs only.” This note travels with any thread you fork from this node.
2

Spawn three parallel threads, one per sub-question

From the framing node, fork three times to create three independent threads:
  • Thread 1 — Labor market effects (Claude Opus): Ask specifically about employment outcomes from UBI pilots. Claude surfaces the Stockton SEED program (+12 percentage points in full-time employment) and the Finland pilot, which showed no drop in labor participation.
  • Thread 2 — Healthcare outcomes (Gemini 2.5 Pro): Direct Gemini to healthcare data. It returns the Manitoba Mincome study showing an 8.5% drop in hospitalizations and the Eastern Band of Cherokee Nation casino dividend study.
  • Thread 3 — Education spillover (GPT-5): Point GPT-5 at schooling data. It covers the Cherokee study’s school completion rates and broader education spillover findings.
Each thread inherits context from the framing node but stays focused on its own domain.
You can run all three threads concurrently. While Thread 1 is processing, start Thread 2 and Thread 3 — the canvas holds them all without waiting.
3

Review findings side by side on the canvas

Arrange the three thread nodes as neighbors. You can now read labor, healthcare, and education outcomes at the same time without scrolling back and forth through one chat. Look for intersections — the Cherokee study appears in both Thread 2 and Thread 3, which signals a high-value primary source worth investigating further.
4

Merge into a synthesis node

Once you have findings you trust, create a merge node. In your own words, pull the key data points from each thread into a unified summary. Because you authored the merge, you control which findings make the cut and how they’re framed — the model doesn’t blend the threads for you without your input.
Use a sticky note on the merge node to pin your synthesis constraints: “cite specific studies and sample sizes, no editorializing.” This keeps any follow-up questions on the merge thread grounded.

​
Tips for research canvases

Each thread node can carry a sticky note that acts as a standing constraint for that thread. For a labor market thread, pin a note that says “discuss employment rate changes only, not income or health outcomes.” The model will stay on task across multiple follow-up turns in that thread.
Not every thread needs the same model. Use a reasoning-heavy model like Claude Opus for threads that require nuanced interpretation of complex studies. Use a model with strong data retrieval like Gemini 2.5 Pro for threads where you need specific statistics. Switching is per-message, so you can also escalate within a thread if a question gets harder.
When a thread surfaces something unexpected — a study you hadn’t heard of, a conflicting data point — fork from that message to investigate it separately. The fork carries full context from the parent thread. Your main thread continues uninterrupted.
Give each thread node a clear label (“Labor — SEED & Finland”, “Healthcare — Mincome”). When you have six or eight nodes on a canvas, names make navigation instant.