Logseq

Last reviewed Aug 26, 2026 · 886 words

There is no Logseq server to install. Logseq is a desktop app (Windows, macOS, Linux) plus mobile apps that read and write a folder of plain Markdown files, and the only thing you self-host is the mechanism that moves that folder between devices. That sounds like a downside for a site about self-hosting; it is the best property Logseq has. Your notes are text files in a directory you control, readable by grep, versionable by git, and syncable by whatever you already run. A 10,000-page graph is a few megabytes.

What the folder looks like, and why it matters

Create a graph and Logseq makes a directory with pages/, journals/, assets/, and logseq/. Each page is a .md file (or .org if you set :preferred-format :org in logseq/config.edn); each day's journal is journals/2026_08_26.md. Blocks are bullet points, nesting is indentation, links are [[Page Name]], and a linked page that does not exist yet is created on the first click. Block references write an id:: property line under the referenced block, which is the one piece of Logseq-specific syntax that ends up in your files.

Because the format is so plain, moving out is trivial. Obsidian can open a Logseq folder with only minor cleanup, which is why the notes apps you own post rates Logseq near the top for exit cost. It also means every other tool on your machine can read your notes: I have a cron job that greps my journals for TODO and posts a count to my dashboard, which no hosted note app could offer.

Outliner, not editor: the distinction that decides if you'll like it

Logseq is an outliner. Every line is a block; every block can be collapsed, moved, referenced, and embedded elsewhere; and the daily journal is the front door. If you think in bullet points and want to write during the day and file later via links, it fits. If you write long prose documents with headings and paragraphs, you will fight it, and Joplin or Trilium are better shapes for that. The graph view is pretty and I open it about twice a year; the linked-references panel at the bottom of every page, showing every block anywhere that mentions it, is the feature that actually delivers "networked thought".

Queries are the sleeper feature. {{query (and (task TODO) (between -7d today))}} on any page gives a live list of every open task from the last week. Advanced queries take Datalog against the whole graph, which is more power than most people need and exactly the power a few people want.

Sync: three options, ranked

Logseq Sync is the project's own hosted, end-to-end-encrypted service. It works, it is the easiest option for phones, and it is the one that is not self-hosted.

Git is the option built into the desktop app: Settings → Version control enables auto-commit on an interval, and you push to any remote you run, such as a Gitea or Forgejo instance. Full history of every note, free, but the mobile apps do not pull it, so this is the choice for desktop-only users or people willing to run a separate git client on their phone.

Syncthing is what I use and what most self-hosters land on. It syncs the graph folder between devices peer-to-peer with no server in the middle, and the Android client works with the Logseq Android app pointed at the same folder. Two rules keep it painless. First, tell Syncthing to ignore the noise, in .stignore at the graph root:

logseq/bak/
logseq/.recycle/
.git/

Second, never have Logseq open on two devices editing the same page at once. Syncthing resolves that by writing a .sync-conflict file, which Logseq then shows as a duplicate page, and untangling it by hand is the only recurring annoyance I have had in three years.

Where it strains

The catalogue's "performance dips on huge graphs" is real. Startup re-indexes the whole graph, and past roughly 20,000 pages or a few years of dense journals (my estimate, not a benchmark), launch on a laptop goes from 2 seconds to 20 and search gets laggy. The file-based architecture also has no server-side search or web access; you cannot open your notes from a browser on someone else's machine. The Logseq team has spent a long time building a database-backed version to address the performance side; check the project's current status before assuming the file-based behaviour described here is what a fresh install gives you.

What I'd do

Install the desktop app, keep the default Markdown format, put the graph in a Syncthing folder with the ignore file above, and enable git auto-commit on your main machine for history. Use the journal every day and let pages accrete from links. If you need phone access, run Syncthing on the phone too. If you want a web UI, multi-user editing, or a server that is the source of truth, Logseq is the wrong tool and Trilium is the right one; for a single person who wants Roam without the rent, Logseq is the answer.