Mindmaps, Nodes & Claims

The mindmap is your persistent research workspace. It doesn't reset between sessions — it accumulates and evolves as you go, reflecting how your understanding of a topic develops over time.

Tree structure and levels of abstraction

A mindmap has a tree structure. A project can have many mindmaps, each with its own organization and agent conversation history. The hierarchy is a first-class tool, not just decoration.

You can organise a mindmap however you like, but the tree becomes most useful when sibling nodes sit at the same level of abstraction — grouping like things together makes gaps and missing angles immediately visible. Another approach is to use the structure to represent a paper outline (Introduction → Methods → Discussion), or to arrange evidence by category, or any other attribute or scheme that fits your research.

  • The agent is designed to adapt to the structure you provide. It can either produce continuous prose (see below) or systematically locate and arrange evidence drawn from your corpus.
  • Transforming a collection of draft nodes into a coherent argument is a common workflow. You might sketch your ideas in a rough structure and then ask the agent to produce a more polished version, rearranging nodes, filling in gaps, and locating citations to support your claims.

Citations make nodes into claims

Citations attach passages directly from your sources to nodes. A node with citations is effectively a claim backed by your sources: the content says something, the citations show where it comes from. Clicking a citation opens the source PDF at the exact passage.

There are two types of citations:

  • Indexed citations — passages retrieved from your KB by the agent or via the RAG Queries panel.
  • Manual citations — free form text that have a citation badge but aren't linked to any source.

Manual citation have an orange badge, while indexed citations have a green badge. Both types of citations can be attached to any node, and a single node can have multiple citations of either type.

What the agent writes

When you give the agent an instruction, it produces nodes that match the natural structure of the task. For argumentative or explanatory work it writes continuous prose — not just bullet points. That means claim nodes, transitional glue text, summaries, and reader-guidance sentences that connect ideas across the map, all properly interleaved. Every substantive claim carries its citations automatically.

When sources disagree, the agent creates sibling nodes to explore various viewpoints rather than averaging the disagreement away. Each viewpoint gets its own node and its own citation chain. You can evaluate them independently. Expand on any of them if needed, etc.

Editing and reorganising

The mindmap is fully editable at any time:

  • Edit node text — rewrite, or replace agent-generated prose with your own.
  • Add nodes manually anywhere in the tree.
  • Move nodes using the keyboard to restructure the tree freely.
  • Delete nodes — recoverable from version history (undo/redo) if you change your mind.
  • Pin citations to any node from the semantic search panel or labeled items panel.

For larger structural changes, you can ask the agent directly: "Move all the methodology nodes under a new parent", "Merge these two siblings", or "Rearrange these children by chronological order".

Undo, redo, and diff view

Every change — whether you made it or the agent did — is tracked. Full undo/redo is available. Each agent edits also exposes a diff view so you can see exactly what was added, modified. See Version History & Provenance.

Giving the agent context

Nodes drive what the agent knows about. When you type a message, you can attach any node (along with its full subtree) as context so the agent understands which part of the map you're talking about.

See Agent Context for the full explanation of how selection and pinning context items scope the agent's view.

See also

What's next