Your Notes App Is a Supply Chain: Obsidian vs Logseq After the Postman Fiasco
After Postman's plugin scandal, this practical comparison of Obsidian and Logseq focuses on local-first knowledge management: licensing, file formats, unsandboxed plugins, sync risks, and git-based portability.
Your Notes App Is a Supply Chain: Obsidian vs Logseq After the Postman Fiasco
The Postman GitHub issue hit 1205 points on Hacker News this week, and the summary in the title is not an exaggeration: users reported that the desktop app silently installed AI-related plugins and, in at least some configurations, attempted to ship API data off the machine. Postman's collections are, for a lot of teams, the closest thing they have to an inventory of every internal endpoint and credential reference they own. That's not a plugin. That's the crown jewels.
I don't use Postman. But I do keep about 4,000 Markdown files in a vault that contains my homelab inventory, half-written runbooks, client notes, and roughly nine years of bad ideas. And the Postman thread made me go look at what my note-taking app was doing on the network.
The answer, for both apps I'm going to compare here, is "not much." That's the whole point.
Why this matters more than the usual tool debate
The Postman story isn't really about Postman. It's about a pattern: desktop app + user account + auto-updating plugin ecosystem + a monetization pressure to bolt on AI features. Any one of those is fine. Stack all four and you get a program that decides on your behalf what leaves your machine.
Your notes are a worse place for that than your API client, honestly. A vault is where you write down the thing you're not sure about yet, the password hint, the client's offhand remark about a reorg. It's unstructured by design, which means no DLP tool will ever flag it, and no one will ever notice it leaving.
So when I say I prefer local-first note tools, I'm not being precious about it. I'm saying the file should be the source of truth and the app should be a viewer. If I can't cat my notes, I don't own them.
There's a second signal in this week's news cycle, and it's easy to miss: Fujifilm is boosting tape production 5x because AI is driving archiving demand. Tape. The least fashionable storage medium on earth, still winning on cost per byte and longevity. Plain text is the tape of note-taking. A folder of .md files will still be readable when Obsidian's company has pivoted to something else, and grep will still be there.
The two real options in 2026
Both Obsidian and Logseq are local-first. Both store files on your disk. Both have mobile apps and sync stories. The similarities end about there.
Obsidian is a proprietary-but-free-for-personal-use editor over a plain folder of Markdown files. What you see in the file tree is what's on disk. The plugin ecosystem is enormous and completely unsandboxed — a community plugin runs with full Node.js access to your filesystem. That's the source of its power and its single biggest risk.
Logseq is AGPL-3.0 licensed and outliner-first. Everything is a bullet. Blocks have identity, which means you can reference a specific paragraph from three different places and it stays linked. Files are Markdown or Org-mode, though with enough Logseq-specific property syntax that "it's just Markdown" is doing some work.
The licensing split matters more than people admit. If you care about the difference between a freemium proprietary app and a genuinely open-source one — and after the last couple of years of license rug-pulls, plenty of us do — the shifts in open source licensing are worth tracking closely. Obsidian has never been open source and has never claimed otherwise, which is at least honest. Logseq's AGPL means you can fork it if the maintainers walk away, and given how long the database version has been in development, that's not a theoretical concern.
The comparison, honestly
| Obsidian | Logseq | |
|---|---|---|
| License | Proprietary, free for personal use | AGPL-3.0 |
| On-disk format | Plain Markdown, folder tree | Markdown/Org with block syntax; DB version changes this |
| Mental model | Files first, links on top | Outliner, blocks first |
| Plugin API | Huge, unsandboxed, full filesystem + Node access | Smaller, sandboxed iframe model |
| Official sync | Paid (Sync), or roll your own | Logseq Sync, with a rocky rollout history |
| Mobile | Polished, free | Functional, more friction |
| Best at | Long-form notes, docs, mixed media, tables | Daily logs, task capture, block references |
| Main risk | Plugin supply chain | Development pace, format migration |
| Network chatter | Minimal; plugin-dependent | Minimal |
Both apps pass the "does it phone home" test out of the box, and you can verify that yourself:
# what is the app actually talking to right now?
sudo ss -tp | grep -iE 'obsidian|logseq'
# or, for a quick list of established connections
sudo lsof -i -P -n | grep -i obsidian
If you run that and see a steady stream of connections to a telemetry endpoint, you've learned something. Neither of these apps showed me anything alarming. Your plugins might, though — that's the part nobody audits.
The practical setup: git as your sync layer
The single best decision I made was refusing to let either app own my sync. A git repo with a remote on a box I control gives me version history, conflict detection, and an offsite copy in one move. It works identically for an Obsidian vault and a Logseq graph, which means the choice between them stops being a lock-in decision.
#!/usr/bin/env bash
# ~/bin/vault-sync — works for Obsidian vaults and Logseq graphs alike
set -euo pipefail
VAULT="${VAULT:-$HOME/vault}"
REMOTE="${REMOTE:-origin}"
cd "$VAULT"
# nothing changed, nothing to do
if [[ -z "$(git status --porcelain)" ]]; then
exit 0
fi
git add -A
git commit -m "vault: $(date -Iseconds) on $(hostname -s)"
git pull --rebase --autostash "$REMOTE" "$(git rev-parse --abbrev-ref HEAD)"
git push "$REMOTE" HEAD
Run it from cron every fifteen minutes:
*/15 * * * * /home/kutak/bin/vault-sync >> /var/log/vault-sync.log 2>&1
Point the remote at a Gitea or Forgejo instance on your own hardware — I've written about standing up that kind of stack in a homelab and it's a couple of hours of work that pays for itself the first time a sync conflict eats an afternoon. If you'd rather not run the git server yourself, a bare repo on any VPS plus Syncthing for the LAN case covers it. The point is that the sync layer is a dumb pipe, not a service with an account.
Once your notes are just files in a directory, the app genuinely doesn't matter. You can search across everything without either editor:
# find every note touching a topic, regardless of which app wrote it
rg -l --type md -e 'homelab|proxmox' ~/vault | head -20
# what did I write this week?
rg --type md -l --sort modified -e '.' ~/vault \
| xargs -r ls -lt | head
And you can point a local model at the directory instead of shipping it to someone's inference endpoint — which is the sane version of the "AI in your notes" feature that Postman apparently decided to install without asking:
# local model, local files, nothing leaves the box
find ~/vault -name '*.md' -newermt '7 days ago' -print0 \
| xargs -0 cat \
| ollama run qwen3:8b "Summarize the open questions from this week's notes."
That's a Qwen model running on your own hardware, reading files that never left it. Given how much of this week's HN front page was about model weights and local inference, it's worth noting that the local-first note-taking crowd has been accidentally building the ideal substrate for this for years.
Which one I actually use, and where it hurts
I've moved between these twice. Right now I'm on Obsidian, mostly because I write long-form documents and Logseq's outliner fights me when I want a 600-word section without a bullet in sight. Obsidian's file tree maps to how I think. Canvas is genuinely useful for mapping out a network topology before I write it up.
But I'll be honest about the caveat: Obsidian's plugin model is a loaded gun pointed at your vault. Every plugin you install is arbitrary code with full filesystem access, running inside an app that auto-updates. I run maybe six plugins, I read the source of anything new, and I keep the vault in git specifically so I can git diff my way out of a bad plugin. That's a lot of paranoia to ask of a note-taking app. Logseq's sandboxed plugin model is the better design and I wish Obsidian would copy it.
Logseq is the better tool for daily notes and task capture. Block references are genuinely superior — being able to embed the same paragraph in a project page and a journal entry without duplication is a feature Obsidian still fakes with plugins. If your primary use is a running log of what you did and what you need to do, Logseq wins. If your primary use is documents you'll revise for months, Obsidian wins.
The thing that keeps me from recommending Logseq outright is the database version. It's been coming for a very long time, it changes the storage format, and the migration path has shifted more than once. The community has been patient. Patience has limits. If you're starting fresh today and want a ten-year bet, that's a real risk factor, and it's the kind of thing that pragmatic automation thinking should account for: don't build a workflow on a format that's mid-migration.
What to do this week
Pick whichever app matches your writing style — outliner if you think in bullets, file tree if you think in documents. It genuinely doesn't matter that much, and anyone who tells you otherwise is selling something.
Then do these three things, in this order:
- Move your vault into git today. One command, and it's the difference between "I lost a week of notes" and "I reverted a bad commit."
- Audit your plugins. Count them. For each one, ask whether you've read the source or trust the author's identity. If the answer is no for more than a handful, cut some. The Postman thread is a preview of what an unnoticed plugin can do.
- Verify the network. Run
ss -tp | grep -i obsidianwhile you work. Know what your tools talk to. It takes thirty seconds and it's the only way you'll ever notice a change.
The app is a viewer. The files are the thing. Keep it that way and you can switch tools whenever the next one gets acquired, enshittified, or decides to install an AI agent you didn't ask for.
Recommended tools & services
Disclosure: some links below are affiliate links. If you buy through them I may earn a commission at no extra cost to you. I only list tools I actually run.
| Tool | Why it is here | |
|---|---|---|
| Amazon - SSDs & NVMe | Boot drives and storage for your homelab | Visit |
| Amazon - Mini PCs | Low-power always-on nodes (ProDesk, EliteDesk, N100 boxes) | Visit |
| Amazon - PC Components | CPUs, RAM and cases for a build | Visit |
| Amazon - GPUs | Consumer GPUs for local inference (16GB VRAM is the sweet spot) | Visit |
| Amazon - RAM & Memory | More RAM is the cheapest upgrade for VMs and containers | Visit |
| Wasabi | Flat-rate hot storage, no egress fees | Visit |