The Interface

A short tour of the app so you know where everything lives.

The sidebar

The left sidebar switches between the main areas of the app:

  • Projects — Create, edit, and attach knowledge bases to projects. Mindmaps, item labels, and information retrieval (RAG queries) are scoped per project.
  • Knowledge Bases — Your indexed libraries, independent of any one project. A single knowledge base can be attached to multiple projects.
  • Project Selection — Switch between projects.
  • Project Contextual Tools — Project-scoped labeled items and RAG queries.
  • Labeled Items — Your theme analysis labels, which can be applied to any RAG query result. Labeled items can be viewed, searched, edited, and managed here.
  • RAG Queries — An interface for semantic searches. Use it to discover information and label items. A similar interface is also available in the mindmap view.
  • Mindmaps — create, edit, and view your mindmaps, which are the home for your research notes and claims.

The mindmap view

Opening a mindmap drops you into its mindmap canvas with several panels arranged around it.

Mindmap canvas (center)

The interactive tree of nodes.

  • Sibling nodes under the same parent often represent the same level of abstraction within a topic. For example, a parent node "Leading theories of Anthropic Reasoning" might have child nodes for each leading theory, with the next level holding the details of each.
  • The mindmap agent uses this structure to place information at the right level of abstraction. For example, if you ask for "an overview of leading theories", the agent will create sibling nodes for each theory. If you then ask to "expand the second theory", it will add child nodes under that theory's node.
  • Each node holds content and may have citations attached to it.
  • Citation-backed nodes show a badge inline — green for indexed citations, orange for manually assigned ones.
  • Clicking the badge opens the Indexed Citations tab or the Manual Citations tab in the right sidebar.
  • The Indexed Citations tab shows the full citation and lets you view the source document or jump directly to the relevant section in the built-in PDF reader.

You can pan, zoom, select single or multiple nodes, expand/collapse subtrees, arrange nodes by moving them around, and edit their content manually. All actions are accessible via keyboard shortcuts and the right-click context menu.

The right sidebar has multiple tabs that allow you to interact with the agent and view or edit contextual information based on the currently selected node.

Chat Tab

The Chat Tab lists all past conversations with the agent. Clicking a conversation opens it, where you can read the history and add follow-up instructions. Typing a message in the list view starts a new conversation. When the agent is running, the panel streams its progress in real time.

Node Properties Tab

The Node Properties Tab shows the content of the currently selected node and lets you edit it.

Indexed Citations Tab

The Indexed Citations Tab shows the citations that are attached to the currently selected node and lets you view their details, including the source document and a link to open the relevant section in the PDF reader. The citation can be expanded to show the full original text and metadata.

Manual Citations Tab

The Manual Citations Tab shows the citations that you have manually attached to the currently selected node. You can view their details and manage them from this tab.

RAG Query Tab

The RAG Query Tab is where you can run semantic searches against the knowledge bases attached to the project. You can view results ranked by relevance, label them, and attach citations directly to the selected node from here by pressing the + button on any result. Full citation text can also be copied to the clipboard.

Labeled Items Tab

The Labeled Items Tab shows the citation items that you have labeled in your project. You can search labeled citations by free text, and filter them by specific label(s), view their details, and attach them directly to the selected node.

Agent Chat Panel

The chat message input panel lets you send instructions to the agent. Each message carries node/subtree context and the full conversation history, so the agent can refer back to previous instructions and its own responses.

  • Model selection — Choose the model to use for each message. Agent Bayes is currently optimized for OpenAI models; GPT-5.1 offers strong answers at a reasonable cost.
  • Mode — Controls how the agent classifies your instruction. Auto lets the agent decide based on context and analyzed intent. Ask restricts it to answering questions and retrieval only (no mindmap edits). Research forces it to look up for new factual information in the knowledge bases. Edit restricts it to restructuring or rewriting existing nodes without research.
  • Research Depth — Controls the scope of knowledge base search. Fast runs a single pass and returns up to 10 citations. Normal runs up to 2 passes and returns up to 20. Deep runs up to 3 passes and returns up to 40. Auto follows the same defaults as Normal unless the agent determines a broader sweep is needed.

The agent always receives the full conversation history alongside the node context you provide. There are two types of context nodes:

  • Selected nodes — Nodes selected on the canvas are passed as previews. The agent reads them and decides whether to load their full content or search for additional context in the mindmap.
  • Pinned nodes — Always provided as full subtree context, regardless of what the agent decides. Press A to pin a selected node, or double-click it. Press X to unpin.

For a deeper explanation of how context scoping works, see Agent Context.

Additionally, you are able to limit the research performed by the agent to a group of documents that you tag by providing a list of tags with your message.

What's next