Your AI re-reads everything, every single question. The fix: a map of your files
A knowledge graph is a map of everything you have - documents, PDFs, notes, screenshots - built once, with the links between them. Your AI consults the map instead of re-reading it all: cheaper, faster, and a memory that sticks.
What it actually is
When you ask your AI “find what I decided about pricing” or “give me the status on this client”, here’s what happens backstage: it digs. It opens files, skims them, opens more, starts over. Like someone leafing through the entire binder, page by page, for every single question. It works - but it’s slow, it’s expensive, and by the next conversation it has forgotten everything: it leafs through again.
A knowledge graph is the other approach: you build a map, once. Every document, every note, every screenshot becomes a dot on the map, and the tool draws the links between them: this quote points to that client, this procedure touches that offer, this accounting folder connects to your online community. Then, when you ask a question, your AI looks at the map instead of re-reading everything.
That’s exactly what Graphify does - a free, open-source plugin for Claude Code: you point it at your folders, it builds the graph - and it keeps it.
The real lever: three things digging will never give you
A memory that sticks. The map is persistent: it survives from one conversation to the next. You stop setting the scene every session; your AI already “knows” the terrain.
A cost that melts. Every page the AI reads is paid in tokens - the unit that measures what the AI reads and writes, and therefore what you pay. Consulting a map costs a fraction of re-reading everything: the project claims up to 70 times fewer tokens per question on large sets of files. Even taking the number with caution, the order of magnitude is real: less pointless reading, faster answers, a lighter bill.
Links you never saw. This is the most surprising effect: the map connects things you had never put side by side yourself. That operations procedure that touches your revenue directly, that old note that answers a current problem. These aren’t gimmicks: they’re your information, finally laid out together.
How it goes, concretely
- Install the plugin: the install command is on the official repo (link at the bottom), you paste it into Claude Code, done.
- Run the
/graphifycommand and pick the folders to map. There is a file limit, so you choose: your living folders (clients, offers, procedures), not the 2019 archive. - Ask your questions as usual. The difference isn’t something you see, it’s something you feel: more direct answers, context already there.
- Visual bonus: if you want to see the map, the free app Obsidian can display the graph - a constellation of your documents, links drawing themselves. Beyond the wow factor, it’s a genuine way to rediscover what you own.
For the owner - and for the employee
If you run the business: your history - quotes, client threads, procedures, decisions - is an asset sleeping in folders. The map wakes it up: your AI can finally lean on everything you’ve already produced, without you spending your life re-explaining it. It’s the same move as with skills: your knowledge, made reusable.
If you’re an employee: notice what the tool does not do. It maps, it doesn’t decide. Choosing which folders deserve to be on the map, spotting which revealed links matter and which are noise, deciding what to do about them - that’s judgment, and that’s you. The person who can read the map is worth more than the map.
Let’s be honest about the limits
- It’s a plugin you’re letting in. Same as any skill: you check where it comes from and what it asks for before you plug it in. The full reflex is in the skills article.
- A map of messy folders is still a mess - nicely drawn. If your files are empty or stale, the graph will be too. The sorting at the door is on you.
- The map ages. Your files move, the map has to follow: plan to rebuild it now and then, or your AI ends up consulting a floor plan from before the renovation.
- The file limit is not a footnote. It forces you to choose - which is rather healthy: start small, one or two folders that matter, widen if it pays.
- The “70 times fewer” comes from the project itself, measured on large volumes. Your mileage will vary. The real gain shows up in use, not on the brochure.
The goal isn’t a pretty constellation to show off. It’s your AI no longer starting from zero - and the time you spent re-explaining your own work coming back into your day.
The official sources
- Graphify - the site: graphify.net
- Graphify - the official repo: github.com/safishamsi/graphify
- Claude Code - the docs: code.claude.com/docs
- Claude Code: claude.com/claude-code
- Obsidian (to see the graph): obsidian.md