From scattered notes to a connected knowledge graph: a step-by-step migration

*May 13, 2026 · 10 min read*

Most people don't start with one note-taking app. They start with five or six: a Notion workspace that grew unwieldy, an Obsidian vault for personal writing, Apple Notes for quick captures, browser bookmarks, a Readwise account, a stack of Word documents from school or work. Each of these tools makes sense in isolation. Together they make a fragmented archive that nothing — including you — can search across.

This is the playbook for consolidating that mess into a single connected knowledge graph without losing anything important. It's structured around the most common starting points (Notion, Obsidian, Apple Notes, scattered markdown) and uses MindWiki as the destination, because the format (plain markdown + frontmatter + wikilinks) is the most portable target.

TL;DR

Five steps in order: (1) inventory every place notes currently live, (2) export each one to markdown, (3) stage the exports in a single folder, (4) link related notes once they're in one place using wikilinks, (5) let the AI surface the connections you missed. The hardest step is psychological — accepting that some old notes won't survive the move, and that's fine. The whole point of a connected knowledge graph is that the surviving notes finally compound. MindWiki gives you a markdown vault with a knowledge graph view and AI access from day one; the migration steps below land you in it.

Step 1 — inventory

Before exporting anything, write down where your notes actually live today. A typical list looks like:

A Notion workspace (work and personal mixed).
An Obsidian vault (personal writing, maybe Zettelkasten attempts).
Apple Notes (quick captures, voice memos transcribed).
Browser bookmarks (saved articles you mean to read).
A Readwise account (book highlights, article highlights, Kindle highlights).
A folder of Word documents from school/work.
Email drafts and labels.
Maybe Evernote, Bear, Roam, or something even older.

Some of these you'll migrate fully. Some you'll harvest selectively and abandon. The point of writing the list is so you don't forget a corner three months later.

Step 2 — export each source to markdown

Markdown is the lowest common denominator. Every tool worth migrating from supports it. Specific paths:

Notion

Workspace → Settings → Export → Markdown & CSV. You'll get a zip of folders that mirrors your sidebar. Notion databases export as CSVs alongside the markdown. The markdown is decent; embeds (Figma, Loom) become URLs.

Obsidian

Already markdown. Skip to step 3 — just copy the vault folder.

Apple Notes

Apple Notes is the hardest. You have three options:

Manual copy/paste for the handful of notes you care about.
Notes Exporter (third-party tool) to bulk-extract.
AppleScript export if you're comfortable with that route.

Honest advice: don't try to bulk-migrate Apple Notes. Most of it isn't worth keeping. Hand-pick the few real ones.

Readwise

Settings → Export → Markdown. You'll get a Books/ folder with one file per source, headlines as #, highlights as bullets. Drop the whole folder into your staging area.

Word documents

Use Pandoc: pandoc input.docx -o output.md. Lossy on tables and embedded images; usually fine on the body text.

Browser bookmarks / read-later lists

These rarely belong in a knowledge graph. Export to CSV from your browser, scan once, copy the few you'd actually re-read into a single reading-list.md page in your staging area, then archive the rest.

Step 3 — stage everything in one folder

Create a new folder named something neutral (vault/) and move every export into it under a structure that reflects the source:

vault/
  capture/           ← anything not yet processed
  notion-export/
  obsidian-vault/    (if you had one)
  readwise/
  word-docs/
  reading-list.md

The temptation here is to start re-organizing into a "real" structure. Don't. The structure you want will emerge from the linking work in step 4 — not from a plan made before you've seen the material in one place.

Once it's staged, point MindWiki at it: install the macOS app, sign in, and choose this folder as your vault path. It'll start syncing immediately. The same vault is then accessible from the web app. Or, if you want to start cloud-first: drop the markdown into capture via the web Quick Capture or the API.

This is the step nobody talks about and the one that makes everything compound. Open a few notes and start dropping [[wikilinks]] whenever you mention an idea that lives somewhere else in the vault.

Practical heuristic for the first month:

When you mention a person, link to their page (create it if it doesn't exist).
When you mention a concept, link to the page (create a ghost link if missing).
When you mention a source (book, paper, talk), link to the source page.
When you mention a project, link to the project page.

You're not trying to link every word. You're trying to make sure that re-encountering one note pulls in the others naturally. The backlinks panel makes this visible from the other side.

Step 5 — let AI surface the connections you missed

After you've done some manual linking, the AI layer earns its keep. Connect Claude (Pro), ChatGPT (Plus), or any other MCP-aware client to your MindWiki vault via the setup guide, and run prompts like:

> "Find pages in my vault that talk about [topic] but aren't linked together yet. Suggest wikilinks to add."

> "Show me pages I haven't touched in 90 days that are still relevant to my active projects."

> "Cluster my slips/ folder by theme and propose a folder structure."

For MindWiki Pro users, the scheduled automations — Auto-Linker, Weekly Classifier, Pattern Detection, Monthly Summary — do versions of this on a schedule. Auto-Linker proposes wikilinks daily; Pattern Detection surfaces cross-domain clusters weekly. You stay the editor (the first three are proposal-first); Monthly Summary writes one additive month-end page.

What to do about formatting weirdness from exports

Notion exports leave some artifacts: empty H1s, escaped underscores, broken image links. A few find-and-replace passes clean most of it:

Replace \_ with _.
Remove leading # from the first line if it duplicates the filename.
Convert Notion's <aside> blocks (which become broken HTML in the export) into MindWiki callouts: > [!note] ....
Image embeds: if you exported with images, they'll be in an Images/ subfolder. Move them under _assets/ (the MindWiki convention) and update the markdown image paths with a single find-replace.

Word-doc exports via Pandoc rarely need cleanup beyond Pandoc's defaults. Readwise exports are usually clean.

What you'll abandon along the way (and that's fine)

Almost everyone, two months in, realizes:

Half their old notes were not worth migrating.
The bookmark dump never gets read.
Apple Notes had three pages worth keeping out of a hundred.
Notion's nested-database labyrinth was hiding most of its content from search.

Don't migrate everything. Migrate the highest-value 20%, write proper slips for what you find, and let the rest live in an archive folder you can search but don't browse.

The architecture you end up with

After the migration, your knowledge surface looks like:

One markdown vault, on your Mac and on the web, syncing in seconds.
One graph view showing the linked structure.
One inbox (capture/) for new material.
One AI connection point — Claude.ai, Claude Desktop, Claude Code, ChatGPT, Codex all reading the same vault via MCP.
One backup/export path — your vault is just files, export anytime.

That's the promise of "connected notes." Most people never get there because they're stuck managing five fragmented stacks instead of one. The migration above takes a weekend if you accept that abandonment is part of the work.

Why MindWiki is a good destination

Plain markdown so you're never locked in again.
Wikilinks + backlinks as first-class, not a plugin.
Knowledge graph view on the macOS app.
Hybrid search (keyword + vector) so you can find half-remembered notes.
MCP server at https://api.mindwiki.io/mcp so every AI client reads the same vault.
Free tier for the vault, search, graph, and sync. Pro adds MCP and the scheduled automations.