> ## Documentation Index
> Fetch the complete documentation index at: https://docs.node-pad.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Forking

> Branch from any message so you can explore a new path without overwriting the current one.

export const EdgeChip = ({type = 'continuation', children}) => <span className="np-edge-chip" data-type={type}>{children}</span>;

Branch from any message so you can explore a new path without overwriting the current one. Each split appears on the canvas as a <EdgeChip type="fork">fork edge</EdgeChip>.

## When to fork

<Columns cols={2}>
  <Card title="Get a second opinion" icon="scale">
    Ask another model the same question before you commit to a decision.
  </Card>

  <Card title="Chase a tangent" icon="split">
    Explore a side question without polluting the branch you still care about.
  </Card>

  <Card title="Compare prompts" icon="columns-2">
    Send two different follow-ups from the same point and see which framing works better.
  </Card>

  <Card title="Preserve a checkpoint" icon="bookmark">
    Split before a risky change so the original path stays intact.
  </Card>
</Columns>

## How branching works

<Steps>
  <Step title="Pick the source node" icon="mouse-pointer-2">
    Right-click the node that represents the context you want to preserve.
  </Step>

  <Step title="Branch from here" icon="git-branch">
    Choose **Branch from here**. NodePad creates a new draft connected to that source node.
  </Step>

  <Step title="Send a new follow-up" icon="send">
    The new draft starts from the same history up to that point, then diverges from what you send next.
  </Step>

  <Step title="Compare results" icon="columns-2">
    Both branches stay visible on the canvas, so you can compare results without losing provenance.
  </Step>
</Steps>

## Patterns that work well

<AccordionGroup>
  <Accordion title="Same question, different model">
    Branch from a user message, keep the wording the same, and route the branch to another model. This is useful when you want a true second read on the same prompt.
  </Accordion>

  <Accordion title="Same branch, tighter scope">
    Branch from a broad response and ask one narrower question in the new branch. The main branch stays high level while the new one goes deep.
  </Accordion>

  <Accordion title="Re-run versus branch">
    Re-run is useful when you want to try the same message again. Branching is better when you want an alternate path you plan to keep around.
  </Accordion>

  <Accordion title="Branch before merge work">
    If you know a branch is going to become source material for another message later, branching early keeps those sources cleaner and easier to reference.
  </Accordion>
</AccordionGroup>

<Note>
  Branches are explicit graph structure. They are not just copied text in a new chat window.
</Note>

## Continue

<Columns cols={2}>
  <Card title="Merging" icon="git-merge" href="/concepts/merging">
    Combine messages from different branches into one new draft.
  </Card>

  <Card title="References" icon="at-sign" href="/concepts/references">
    Cross-link across branches with `@`.
  </Card>
</Columns>