How to highlight text and branch conversations¶

Text highlighting lets you extract specific passages from nodes and respond to them, creating focused sub-conversations on the canvas.

What is highlighting?¶

When you select text within a node and create a highlight:

A new HIGHLIGHT node appears containing just that text
A dashed line connects the original node to the highlight
You can reply to or branch from the highlight independently

This is useful for:

Responding to a specific point in a long message
Extracting key quotes or findings for later reference
Branching the conversation in a new direction from one specific idea

Creating a highlight¶

Method 1: Highlight and branch¶

Select text within any node by clicking and dragging
A tooltip appears showing the selected text
Click the "🌿 Branch" button in the tooltip
Optionally type a reply or command in the input field
Click "Branch" (or press Enter)

A highlight node appears near the viewport center, connected to the source node with a dashed line.

Method 2: Highlight then reply separately¶

Select text and click "🌿 Branch"
Leave the reply field empty
Click "Branch"
The highlight node appears
Select it and type in the chat input to reply

What happens when you branch¶

Without a reply¶

Just creates the highlight node:

Original node: "Quantum computing uses qubits which can exist in superposition..."

Highlight node: "> Quantum computing uses qubits which can exist in superposition"

The > prefix formats it as a blockquote.

With a reply¶

Creates the highlight node and immediately responds to it:

You select: "uses qubits which can exist in superposition"
You type: explain this in simpler terms
Result: Highlight node + AI response node

With a slash command¶

You can use slash commands when branching:

Select: "revenue grew 40% last quarter"
Type: /research what market factors contributed to this growth
Result: Highlight node + Research node

The selected text becomes context for the slash command, helping the AI refine its search or research query.

Working with highlights¶

Highlight nodes are full nodes¶

They work exactly like other nodes:

Reply to them
Select them as context for new messages
Delete them if no longer needed
Move them around the canvas

Dashed connections¶

The dashed line from source → highlight is a HIGHLIGHT edge type. This visually distinguishes highlights from regular replies (solid lines).

Multiple highlights from one node¶

You can create many highlights from the same node:

Research paper node
  ↦ "Methodology section" (highlight)
  ↦ "Key finding #1" (highlight)
  ↦ "Surprising result" (highlight)

Each can spawn its own conversation thread.

Common use cases¶

Discussing research or long documents¶

When working with imported PDFs or research reports:

Skim the full document node
Highlight key findings, one at a time
Ask questions about each highlight
Build a knowledge graph of interconnected insights

Responding to specific points¶

In a long AI response with multiple ideas:

Highlight just the part you want to discuss
Reply to that specific point
The conversation focuses without repeating the entire context

Without highlights:

You: "Can you elaborate on the third paragraph about database indexing?"
AI: [has to re-identify which part you meant]

With highlights:

You: [highlight third paragraph]
You: "Can you elaborate on this?"
AI: [knows exactly what "this" refers to]

Extracting images from nodes¶

When a node contains an image (like a screenshot or diagram):

Right-click the image
Select "Extract as separate node"
A new IMAGE node appears containing just that image
Connected to the original via a HIGHLIGHT edge

This works like text highlighting but for visual content.

Tips for effective highlighting¶

Keep highlights focused¶

✅ Good: Highlight a single idea or finding

> "The model achieved 94% accuracy on the validation set"

❌ Too broad: Highlighting multiple paragraphs

> [giant wall of text spanning 5 different topics]

For large sections, just reply to the full node instead.

Use highlights for non-linear reading¶

When reading a long document:

Don't read top-to-bottom
Scan for interesting parts
Highlight and explore each in depth
Create a web of interconnected insights

Combine with matrix evaluation¶

After highlighting several key points:

Select all highlight nodes
Run /matrix evaluate these findings against novelty, feasibility, and impact

Branch from highlights¶

Highlights can themselves be highlighted:

Research node
  ↦ Highlight A
      ↦ Highlight A.1 (specific sentence within A)
          ↦ Deep dive conversation

This creates nested focus, zooming in on increasingly specific details.

Context propagation¶

When you highlight text and then reply or run a command:

The selected text becomes context¶

If you highlight: "transformer attention mechanism"

Then type: /search how does this work?

The AI refines your query to: "how transformer attention mechanism works"

The vague pronoun "this" is resolved using your highlighted text as context.

Why this matters¶

It enables conversational slash commands:

/search what are alternatives? — alternatives to what? Your highlight!
/research explain the implications — implications of what? Your highlight!

Differences from regular replies¶

Feature	Regular Reply	Highlight + Reply
Creates nodes	1 (reply)	2 (highlight + reply)
Edge type	REPLY (solid)	HIGHLIGHT (dashed)
Source content	Full parent node	Selected text only
Use case	Continue conversation	Focus on specific point

Troubleshooting¶

"🌿 Branch" button doesn't appear¶

Make sure you've selected text (click and drag)
Try selecting text again
Some nodes may not support text selection (check node type)

Highlight appears in wrong location¶

Highlights appear near the viewport center
Move them manually by dragging
Or use auto-layout (🔀 button) to reorganize

Selected text is not visible in highlight node¶

Check that you selected text before clicking Branch
Highlights should show the text with > prefix (blockquote)
If empty, try creating the highlight again

Can't highlight images¶

Images use a different mechanism: right-click → "Extract as separate node"
This creates an IMAGE node, not a HIGHLIGHT node
Connected with the same dashed line style

Advanced patterns¶

Multi-stage highlighting¶

Import a research paper (PDF) - see How to import PDFs for details on drag/drop and /note URL methods
Highlight key sections
For each section, highlight specific sentences
Ask detailed questions about each sentence
Build a deep understanding layer by layer

Highlight + Committee¶

Find a controversial claim in a node
Highlight just that claim
Run /committee is this claim accurate? - see How to use LLM committee for details
Get multiple AI perspectives on that specific assertion

Highlight + Matrix¶

Highlight 5-10 different ideas from various sources
Select all highlights
Run /matrix compare these ideas against originality, feasibility, evidence - see How to use matrix evaluation for details