🧠 Fundamentals
Before installing anything: why giving Claude Code memory changes the game, what a knowledge graph is (no mystery), what Graphify does, and why Obsidian is the right home for all of it. Each term is defined the first time it appears.
progress
↑ The track’s path in one image: documents become entities linked by relationships, grouped into communities, with a god node in the center—and Claude Code consults this map instead of rereading everything.
Trail map
Detailed content
🧠 Why Claude Code needs a second brain
The difference between a session that forgets everything and an agent with searchable external memory.
Claude Code is a agent of AI that runs in your terminal—it reads files, edits code, and runs commands for you. “Memory” here means everything it can keep in mind to respond: what’s in the current conversation plus whatever you point it to.
Without external memory, each session starts from scratch on large projects. Understanding what it "remembers" is the first step toward giving it a persistent brain.
Agent · session · context · external memory.
A context window is how much text the model can read at once (measured in tokens — word fragments). It’s large, but finite: an entire repository rarely fits.
When the content won’t fit, the agent needs to choose what to read. A knowledge map helps it choose well instead of scanning everything blindly.
Token · context window · cost · context selection.
grep is a literal search for a word in the files. It finds where the term appears, but doesn't know what it relates with what or why.
That’s exactly the gap the graph fills: it stores the connections and the meaning, not just the text's location.
Literal vs. semantic search · relationship · "map, not index".
A "second brain" is an organized external knowledge base that you (or an agent) consult. Here, it's the graph + vault that hold a project's understanding.
It’s the course’s central idea: take knowledge out of the ephemeral session and put it somewhere persistent and navigable.
PKM · external memory · persistence · searchable knowledge base.
Claude Code already reads a file CLAUDE.md with instructions and its own memory. It’s great for rules, but it isn’t a map of an entire repository.
To find out where the native tool ends and Graphify+Obsidian begins — they complement each other; they don't compete.
CLAUDE.md · native memory · limits · complement.
Graphify creates the map; Obsidian stores it as navigable Markdown; Claude Code queries it in your project context. Together = a second brain.
It's the roadmap for the entire course. Knowing the destination makes each practical step that follows make sense.
Stack · map → vault → agent · project context.
🕸️ Knowledge graph, demystified
Nodes, edges, communities, and god nodes — the vocabulary that makes the rest of the course obvious.
A graph are points (nodes) connected by lines (edges). Think of subway stations (nodes) and tracks (edges) between them.
It’s the structure Graphify produces. Without this mental picture, the rest becomes jargon.
Node · edge · neighborhood · degree (number of connections).
In Graphify, the nodes are entities (an idea, function, person) and the edges are relationships typed. Each one carries provenance: which file it came from.
The video says "concepts/connections"; the tool says "entities/relations". Knowing both terms avoids confusion in practice.
Entity · typed relationship · provenance · reconciliation (merging aliases).
A community is a group of highly connected nodes. The Leiden is the algorithm that finds these groups automatically.
Communities become the project’s “themes”—and, in Obsidian, named groups on the canvas. That’s how the graph becomes navigable.
Community · clustering · Leiden · theme/topic.
A god node is the most connected entity — the hub that touches many others. Example: "Context Window" pulling in dozens of related concepts.
These are the best starting points for asking the agent — and GRAPH_REPORT.md lists them for you.
God node · high degree · hub · entry point.
RAG is "searching for passages and giving them to the model." GraphRAG does this for a graph: beyond the passages, it provides the relationships between them.
It’s the difference between giving the agent a stack of pages and giving it an annotated map. A map gets you faster, better answers.
RAG · GraphRAG · retrieval · structured context.
With the map, Claude Code knows what's close to what. Ask about subagents → it finds "agent teams," "separate context," and so on, without scanning everything.
It’s the selling point of the entire stack — better answers in large repositories.
Neighborhood · navigation · contextual answers · efficiency.
⚙️ Graphify under the hood
What the tool does, how it extracts knowledge, what it produces—and the limitation Obsidian solves.
Graphify is a CLI (terminal program, the package graphifyy) that also installs a skill (an instruction file) inside Claude Code.
The two sides do different things: the skill (/graphify) has Obsidian mode; the headless CLI has extract/query. You'll use both.
CLI · skill · graphifyy · slash command.
For code, Graphify reads the structure (the AST, the code “tree,” via tree-sitter)—fast and without an API key. For documents, use a LLM to extract the meaning.
Decide whether you need an API key and how "literal" or "semantic" the extraction will be.
AST · tree-sitter · semantic extraction · LLM.
Everything goes into one folder graphify-out/: o graph.json (the truth), the graph.html (interactive visual) and the GRAPH_REPORT.md (audit with suggested questions).
Knowing which file is which prevents panic on the first run and shows you where to explore.
graph.json · graph.html · GRAPH_REPORT.md · cache.
Graphify points both to a code base how much for one document corpus (PDFs, markdown, etc.). In the course, we use documents: the Claude Code docs.
The choice changes the type of graph and whether you take it to Obsidian or keep it in Graphify itself (the subject of Track 3).
Code base · corpus · markdown · PDF.
The skill has --obsidian (one Markdown file per node with backlinks + one canvas) and --wiki (articles by community, Wikipedia-style).
It's the video's secret sauce: turning the graph into a vault that Obsidian (and the agent) understands.
--obsidian · --wiki · graph.canvas · backlink.
On its own, the Graphify graph exists in isolation — it only knows about that corpus. It doesn’t connect to the rest of what you store.
That’s exactly why we bring it into Obsidian: to fit this knowledge into the broader context of your project.
Silo · vacuum · broader context · integration.
🔮 Obsidian, the vault that becomes memory
What a vault is, why Markdown works well with the agent, and how graph.json becomes a navigable brain.
O Obsidian is a note-taking app. A vault is simply a folder of files markdown (plain text with formatting) that it opens.
Since it’s just a folder of files, Claude Code can read everything too—there’s no proprietary format in the way.
Obsidian · vault · markdown · local folder.
A wikilink é [[nome-da-nota]]: links one note to another. Obsidian creates the backlinks (who points here) by itself.
This is how the graph’s edges become navigable links. Each Graphify relationship becomes a [[...]].
Wikilink · backlink · note · bidirectional link.
Obsidian has a "graph view"—but it's only a drawing of the links between markdown notes. It's not Graphify's knowledge graph; it's a visual representation of it.
The video warns about this: "it’s not exactly a knowledge graph, it’s a bunch of connected markdown files". Knowing the difference prevents false expectations.
Graph view · note links · representation · ≠ original graph.
Markdown is plain text. The agent reads it, runs grep, follows the [[links]] — no binary layer in between.
That's why the vault is the ideal home: the same file works for you to read in Obsidian and for Claude Code to consult.
Plain text · human- and machine-readable · no lock-in.
O Canvas is a visual canvas; the plugins extend Obsidian. Graphify generates a graph.canvas with the communities already grouped.
Many people want the stack just for Obsidian's infrastructure (UI, add-ons). Knowing what's available helps you decide.
Canvas · plugin · UI · add-ons.
The Obsidian export reads the graph.json and writes one Markdown file per node. It’s the bridge between the graph world and the notes world.
Completes Track 1: you understand the two shores and the bridge. Track 2 crosses that bridge in practice.
graph.json → .md · bridge · regeneration · practical next step.