The connections are
the knowledge.

A mind map is a tree with one centre and branches radiating outward, which makes it excellent for exploring one topic quickly and unable to express a relationship between two branches. A knowledge graph is a network: any node can connect to any other, connections carry labels, and there is no privileged centre. In practice mind maps are for generating and graphs are for organising what you have generated. Many projects use both, starting with a mind map when the subject is new and moving to a graph once there is enough material that the cross-connections matter more than the branching structure. On an infinite canvas you can keep both on the same board and let one grow out of the other.

Start with your material rather than with a schema, because a schema designed in advance is almost always wrong and it stops you before you begin. List the entities that keep coming up, put each on a node, and then do the actual work, which is connecting pairs and labeling what the connection is. Resist the temptation to connect everything: a graph where every node touches every other carries no information, and the discipline is including an edge only when the relationship does work. Then read the result for absence, which is what the exercise is really for. Nodes with one source, clusters that connect to nothing, and claims with no evidence attached are the findings, and each one is either a gap to fill or a deliberate boundary.

Typed edges, attached sources, and the willingness to leave gaps visible. Untyped graphs are the most common failure and they look impressive: hundreds of nodes and a dense web of lines that all mean the same nonspecific thing, which is that two notes reference each other. That picture cannot answer a question. A graph where the edges say contradicts, replicates, or depends on can answer several, including the one that matters most, which is what breaks if this source turns out to be wrong. The second requirement is that nodes carry their evidence, because a graph you cannot trace back to primary material becomes a set of assertions you have to take on trust from your past self. The third is not smoothing over the holes.

PKM stands for personal knowledge management: the practice of capturing, organising, connecting, and retrieving what you learn so it stays usable rather than accumulating. It covers note-taking, but the note-taking part is the easy half and the one every tool solves. The hard half is connection and return, and the common outcome of a mature PKM system is several hundred well-tagged notes that their owner has not opened since writing them, because search only answers questions you already know to ask. A knowledge graph is one answer to that: store the relationships between ideas explicitly, so the structure itself tells you which claim rests on one source and which two clusters are about the same thing without knowing it. Popular PKM tools include Obsidian, Roam, and Notion, and the practice matters more than the tool.

Not during early access. Storyflow is paid-only right now, with plans from $7.99 a month billed annually, and a Free plan launches before the end of 2026. Anyone a paid member invites to a board joins free today, so a research group or a co-author can view and contribute without paying. If you need a free option now, Obsidian is genuinely free for local use and its graph view is good, with the caveat that its edges are untyped. For a real graph database with a query language, Neo4j has a free tier. Storyflow occupies the informal middle: no schema to design, no query language to learn, spatial arrangement you control by hand, and an AI that produces the first pass.

It can do the extraction and the first arrangement, which is the tedious part, and it should not be trusted with the judgments. Given a body of material, the AI will identify the recurring entities, lay them out, and propose connections, which turns hours of node creation into something you spend your time correcting instead. That correction is not overhead, it is where the understanding happens: deciding that two findings genuinely contradict rather than merely differ, or that one claim depends on another, is the actual intellectual work and it is not delegable. The specific thing to watch for is the AI over-connecting, since a model asked to find relationships will find them, and a graph with too many weak edges is less useful than one with fewer strong ones. Prune aggressively.

On a fixed page, around forty nodes before the edges become spaghetti. On an infinite canvas the practical limit is much higher because you can use space instead of compressing: give each cluster its own region with internal edges drawn inside it, and only draw cross-cluster edges that carry real weight. Beyond a few hundred nodes the right move is not a bigger picture but a layered one, with a top-level map of clusters and detailed graphs in their own regions. The skill throughout is curation rather than completeness. A graph that includes every relationship you could justify hides the dozen that matter, and the point of drawing it was to make those dozen visible.

It is arguably the best use for one, because a literature review is a graph that convention forces into prose. Put each paper on a node with its finding, and label the edges honestly: replicates, contradicts, uses the same method on a different population, was superseded by. Three things surface almost immediately and are painful to find any other way. Clusters of papers that all cite one original study, which tells you the field rests on a narrower base than the citation count suggests. Contradictions nobody has reconciled, which is often where the contribution is. And your own claims that no paper on the board actually supports. Write the prose afterwards, from the graph, and export it as an image or PDF for a supervisor.

Yes, and shared construction is where graphs earn their keep. Invite people onto the canvas and a research group, a team, or a pair of co-authors builds one graph instead of several private ones that quietly disagree with each other, which is the normal state of affairs and is only discovered at a late and expensive moment. Send a view-only link and anyone opens the whole graph in a browser with no account, which makes it possible to hand the structure of an argument to a supervisor or a colleague rather than describing it. Comments attach to the node or edge they concern, so a disagreement about whether two findings actually conflict stays on the edge between them.

A second brain is a whole personal knowledge practice, covering capture, organisation, retrieval, and review, and it is largely about making sure things you encounter are findable later. A knowledge graph is one structure inside that practice, and it addresses the part most second-brain systems handle worst, which is connection. The common experience of a mature note collection is hundreds of well captured, well tagged notes that are never revisited, because capture and retrieval were solved and synthesis was not. A graph is the synthesis step: deciding what relates to what and why, which is slower than capture and is the only part that produces new understanding rather than storage. In Storyflow both can live on the same canvas.