Personal knowledge mapping
Your notes have a shape. Most tools hide it.
Laminar maps your knowledge against the full landscape of human understanding. Capture without friction. Prune with intention. Then see the gaps — the concepts your knowledge reaches for but doesn't yet contain.
The loop
Most PKM tools blur reading, thinking, and reviewing into one undifferentiated activity. Laminar treats them as distinct cognitive modes — each with its own interface, its own purpose, its own moment.
"Laminar flow doesn't mix layers. Three clean streams. No turbulence."
01
Passive · ambient · zero friction
Happens in the background of your reading life. Highlight text on any screen, leave a voice note as you read, or import from Kindle and Readwise. You never leave the source to feed the system. Raw material flows in — unstructured, uncommitted.
Design rule: capture must cost nothing. If it interrupts the learning, it won't happen. The system accepts everything without judgment. Cleaning comes later.
02
Active · deliberate · effortful
A feed of suggested merges, connections, and orphan alerts. You make decisions — merge these two nodes, keep them separate, add a voice note explaining the distinction. You cannot prune what you don't understand. The pruning is the comprehension.
Design rule: pruning is the learning act itself. A graph you've pruned is a graph you own. This is where passive consumption becomes actual knowledge.
03
Revelatory · earned · directional
After pruning, the graph is clean enough to trust. Now Laminar shows you the ghost nodes — concepts your knowledge points toward but doesn't contain. Located on the global skeleton, so you know exactly where they sit in the landscape of the field.
Design rule: gaps are earned, not given. They only become visible after pruning. A noisy graph has false gaps. A clean graph reveals true absence.
Capture
The moment of learning and the moment of capture should be the same moment. Laminar captures from wherever you are — no tab switching, no copy-pasting, no breaking the reading to feed the system.
Screen highlight
Select any text on any screen. Browser extension captures with full source context — the URL, the date, the surrounding paragraph. One click, done.
The capture is the highlight. Nothing more.
Voice note
Tap once, speak your insight. Transcribed and linked to whatever you were reading when you said it. The thought and its context, together.
Your voice, at the moment of understanding.
Import
Readwise, Kindle highlights, Notion pages, Obsidian vaults, plain text, PDFs. Everything you've already annotated, brought into the graph.
Your existing annotations, finally connected.
Write
A minimal editor for synthesis. Write your own understanding of something. Laminar watches as you type and integrates your formulation into the graph alongside your source material.
Your own words as a node, not just a quote.
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks. The Transformer architecture relies entirely on an attention mechanism to draw global dependencies between input and output, dispensing with recurrence and convolutions entirely.✓ Captured to graph
Unlike recurrent models which process tokens sequentially, the Transformer allows for significantly more parallelisation and requires significantly less time to train✓ Captured to graph on large datasets.
Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.
Just captured → graph
2 new nodes: Attention mechanism, Parallelisation in transformers
Connected to: Transformer architecture (existing) · Self-attention (existing)
Ghost detected: Linear algebra foundations →
Prune
A feed of suggested merges, connections, and orphaned nodes. For each one, you decide. Accept, reject, or speak — add a voice note that explains the nuance. That note becomes part of the edge itself.
The algorithm can suggest connections. It cannot decide which ones are meaningful to you. That decision is the comprehension. A graph you haven't pruned is just a collection of things you've read. A pruned graph is what you understand.
"The pruning session generates new material. As you decide whether two nodes belong together, you often discover a third thing you need to capture."
Merge suggestion
Attention mechanism
Vaswani et al. · 4 connections
Self-attention
Karpathy notes · 2 connections
Both appear to describe the same mechanism captured from different sources. Merging would unify 6 connections under one node.
voice note added
"Self-attention is the specific mechanism — attention is broader, keep separate but connect"
See gaps
After pruning, the graph is clean enough to trust. Now Laminar surfaces the ghost nodes — concepts your existing knowledge actively reaches toward but doesn't yet contain.
These aren't random suggestions. They're inferred from your own graph. Your ML notes reference Linear Algebra 14 times without you ever having captured it directly. That's a ghost. That's your next learning direction.
Solid — you own this
Multiple captures, pruned, well-connected. You understand this concept at depth.
Shallow — present, unpruned
Captured but not yet worked through. You've encountered this but not integrated it.
Seed — captured, unconnected
An orphan waiting for context. You captured it before the surrounding concepts existed.
Ghost — your gap
Not captured. Inferred from what surrounds it. This is what to learn next.
Skeleton — the map
Exists in the global knowledge taxonomy. Not a gap — just unexplored territory you may never need.
The skeleton
Without a skeleton, your graph is self-referential — it only finds gaps relative to what you already know. With it, your knowledge has absolute position. You can see not just the gaps between your nodes, but where you sit in the full landscape of a field.
The skeleton comes from Wikipedia's category graph — the most comprehensive, maintained, neutral taxonomy of human knowledge ever assembled. Curated to three levels: ~12 root domains, ~120 major subfields, ~1,200 topics.
Root domains
Mathematics, Sciences, Humanities, Engineering, Arts… Always visible. Global orientation.
Major subfields
Machine Learning, Linear Algebra, Cognitive Science… Where most users locate themselves.
Topics
Transformers, Backpropagation, Attention… Faint until your notes get close. Brighten on proximity.
Your graph
Everything you've captured, pruned, and connected. The brightest layer. Anchored to L3.
You don't start with a blank canvas
When you open Laminar for the first time, the full skeleton is already there. You locate yourself immediately — "I work here and here." That declaration seeds your graph before you've captured a single note.
Why Laminar
Most tools give you your knowledge.
Laminar gives you your knowledge
located within all knowledge.
You can see not just what you know, but where you are in the larger landscape — and therefore which direction to walk next. The difference between a note collection and a map is the difference between memory and navigation.
Laminar flow: smooth, ordered, non-turbulent. The opposite of how most people's notes work. Three clean streams — capture, prune, see gaps — that never mix, never create chaos, and compound over time into something that actually resembles how you think.
Capture
ambient · passive
Prune
deliberate · active
See gaps
revelatory · earned
Laminar
knowledge in flow
Laminar is in private beta. Early access is open to learners who want to help shape the gap detection model and skeleton taxonomy.
Your early graphs will inform how Laminar identifies gaps across different domains and learning styles.