It pulls out the concepts, wires up the ones that travel together, then runs real graph-theory + language math on the result: PageRank crowns the main character, betweenness exposes the hidden bridge holding two topics together, community detection splits it into idea-tribes, and a mood EKG traces the emotional arc. No upload, no account — it all runs right here in your browser.
I fall hard for topics. A two‑hour talk, a dense paper, a rabbit hole at 2am — I want it, all of it. And then, every single time, I hit the same wall: I'm drowning. Forty ideas shouting at once, and not one of them tells me where to start or which ones actually matter. I read it twice and still can't hold the shape of it in my head. My notes pile up and quietly become one more thing I never reopen. The love is real. The overwhelm is worse — that gap between how badly you want to understand something and how lost you feel standing inside it.
This is the tool I wished I'd had. It takes that wall of text and turns the ideas inside it into a map you can see all at once — what everything leans on, which quiet idea holds two halves together, where the mood turns — so you finally know where to begin and what to hold onto. It doesn't think for you. It shows you how the thing actually hangs together, and hands you the map to walk yourself.
And it's honest about how: it strips ~160 filler words, keeps multi‑word ideas whole (so knowledge graph stays one concept, not two), and draws a link between any two ideas that share a sentence — the more often they co‑occur, the stronger the bond. That web is the map. Then the real math begins.
Yes — the exact algorithm Google was built on, the one that ranked the entire web. Here it's pointed at your concepts instead of web pages: importance flows along the links, round after round, until one idea wins. → It tells you what the text is really about — which is often not what the title says.
It counts which idea secretly sits on the path between everything else (Brandes' algorithm — the same idea network scientists use to find the person who holds a whole organisation together). → The quiet connector linking two topics; pull it out and the conversation splits in two. The MVP a reader feels but can't name.
Every concept keeps adopting its neighbours' most common label until the dust settles — no training, no labels, the clusters fall out on their own. → The camps and sub-topics the text organised itself into, coloured and named.
A compact emotion lexicon scores every sentence (flipping the sign on negation), then draws the result as a heartbeat with the peak and the dip called out. → You see where it turned hopeful and where it turned heated.
Tension (questions + disagreement markers), Certainty (how few hedges like "maybe / probably"), Echo (how repetitive). → Confident argument, open debate, or someone circling the same point?
Pick any two ideas and it traces the shortest chain of concepts between them. And if there's no path at all? → The gap itself is the insight — those two ideas never met in this text.
No black box. The whole analysis runs in your browser, from scratch — then Axon rides on top. Here's what it really is, and the full pipeline if you want to go deep.
An MRI for an idea. You feed it words — a note, an article, a whole PDF — and it rebuilds them as a map of how those ideas hold each other up: which one everything leans on, which quiet one bridges two worlds, where the mood turns. Then Axon walks you through that map from first principles until it clicks.
Every number here comes from an algorithm you could read yourself — PageRank, betweenness, community detection, a sentiment arc — no libraries, no guessing. The maths finds the structure; Axon only ever helps you understand it. It never invents the shape. Open to a deep dive whenever you are — the pipeline below is the honest, step-by-step of it.
Split into sentences, drop ~160 stopwords, then score the surviving uni- and bi-grams by frequency × length. The strongest ~34 become nodes — multi-word ideas like knowledge graph survive as one.
Two concepts in the same sentence earn an edge; the more often they co-occur, the thicker and springier it is. That co-occurrence matrix is the graph.
A from-scratch force simulation: every node repels every other (Coulomb), every edge pulls like a spring (Hooke), gravity keeps it centred. The knobs up top are those constants — nudge them and watch it breathe.
Google's PageRank, run on your concepts: importance flows along the links until it settles. The winner is the main character; node size is its score.
Brandes' betweenness counts how many shortest paths run through each node. High score = a connector — pull it and the argument can split in two. The non-obvious MVP.
Label propagation: each node keeps adopting its neighbours' most common label until stable clusters fall out — coloured and named by their strongest member. No training, no labels.
A compact AFINN-style lexicon scores each sentence (flipping on negation) and draws the emotional arc as a heartbeat. Gauges add tension (questions), certainty (few hedges) and echo (repetition).
The moment your map appears, Axon reads it and opens with the deep, simple truth under it — plus one question you won't leave alone. Click any node and it explains that idea from first principles; ask it anything and it answers only from your text, lighting the ideas it leaned on; hit Blind Spot and it adds what the text never mentioned, as gold nodes. Heavy lifts run on Opus, the rest on Haiku — fast and cheap.