Zettlr, Joplin, Anytype and AppFlowy compared

Most comparisons of personal knowledge tools stop at Obsidian, Logseq and SiYuan, which is a reasonable place to stop for a general-purpose note vault; those three are covered in Obsidian vs Logseq vs SiYuan. Three common requirements fall outside what they do well, and each has an obvious answer among the four tools here.

The first is citations. If a note eventually appears in a manuscript with a bibliography attached, you need a reference manager wired into the editor and an export path that produces a real document. Obsidian can be made to do this with plugins; Zettlr does it out of the box.

The second is encryption and sync you control. A vault of plain Markdown is only as private as the folder it sits in, and syncing that folder through a general-purpose cloud drive means the provider holds readable files. Joplin encrypts before anything leaves the device.

The third is structured data: databases, boards, relations between records, several people in one space. That is Notion's territory, and Anytype and AppFlowy are the two serious local-first attempts at it, from opposite directions.

All four are free to run and keep data on your own machine. They differ sharply in what that data physically is, which is the decision that matters most in the long run.

Zettlr is a Markdown editor written for academic and long-form writing. It opens a directory of Markdown files, adds Zettelkasten-style identifiers, internal links and backlinks on top of them, and leaves the files as plain text with YAML frontmatter. Nothing about the storage is unusual, which is the point: the interesting part sits around the editor.

Citations are first class. You point Zettlr at a bibliography exported from Zotero (or any BibTeX or CSL JSON library), and typing a citation key then autocompletes against your library and renders the reference in the preview. Formatting follows a CSL style, so switching a manuscript from APA to Chicago is a change of style file rather than a pass through the document.

Export goes through Pandoc. Zettlr does not implement its own document conversion; it drives Pandoc, so the output formats are Pandoc's: LaTeX, PDF (via a TeX installation), Word, HTML, ODT, EPUB, revealjs slides. A supervisor who wants a .docx and a journal that wants LaTeX are served from the same source file. The tradeoff is that a full PDF export expects Pandoc and a TeX distribution installed, a real setup step on a fresh machine.

What it is not. Zettlr is desktop only, with no official mobile app, and it is a writing tool rather than a life-management system: no databases, no boards, no task engine. If your notes are also your calendar and your project tracker, this is the wrong shape. If they end up as a chapter or a paper, it is the one to try first, and because the files are ordinary Markdown you can keep another editor open on the same folder.

Joplin is one of the oldest actively maintained open-source note applications, and it shows in the least glamorous way possible: it is unusually complete. Desktop apps for Windows, macOS and Linux, mobile apps for Android and iOS, a terminal client, a plugin API, and a web clipper for Firefox and Chrome that saves a page, a selection or a simplified reader version straight into a notebook.

Encryption is the headline feature. Joplin's end-to-end encryption is opt-in and, once enabled, applies to everything before it is synchronised, so the sync target holds ciphertext. That design is what lets the second feature work.

Sync is deliberately backend-agnostic. Rather than requiring a service of its own, Joplin synchronises through Joplin Cloud (the project's own hosted option, which funds development), Nextcloud, Dropbox, OneDrive, WebDAV, S3-compatible storage, or a plain directory you replicate however you like. If your employer already gives you one of those, you have a sync solution, and the provider stores an opaque blob rather than your notes.

The storage nuance worth knowing. Note content is Markdown, but the working store is a local SQLite database rather than a folder of .md files, so grep over a notes directory does not work the way it does with a Markdown vault. Joplin answers this with export (Markdown files, a directory dump, the JEX archive format, PDF and HTML) and with a file-system sync target that does write files out, but if greppable text is a hard requirement, know this before you commit. The same point comes up throughout PKM for developers, where scripting over the notes is often the whole reason for the setup.

What it is not. Joplin is a notebook, not a thinking tool. Notebooks, tags, to-dos and search are all there, and linking exists but is not the centre of gravity: no graph-first workflow, no block references, no daily-note engine unless a plugin supplies one. Pick it when security and portability matter more than the density of connections between notes.

Anytype is the most conceptually different of the four. It does not think in documents. Everything is an object: a note, a task, a person, a book, a bookmark, a meeting. Each object has a type, and types have relations, typed fields that point at other objects. A book has an author relation pointing at a person, and that person shows every book, note and meeting referencing them. Sets and collections query across the graph the way a Notion database view does, except the query runs against your local store. Editing is block based, and spaces separate contexts.

Sync is peer-to-peer. Anytype's Any Sync protocol replicates encrypted data directly between your own devices, and a network of nodes can hold an encrypted copy so devices never online at the same time still converge. No party in the middle can read the content, and the node software is published, so a technically inclined user can run their own.

Say the tradeoff plainly: this is not Markdown on disk. Anytype keeps objects in its own encrypted local store in a binary format, not as files you can open in a text editor, list with ls or search with grep. There is a Markdown export, and the client code is published (under Anytype's own source-available licence rather than one of the standard open-source licences), so it is not a black box. But it is not the same guarantee as a directory of plain text: what you get is encrypted, local, exportable rather than readable without the application.

It is also the fastest-moving of the four. Anytype has changed its data model, its sharing model and its commercial model more than once as it has matured, and further change is reasonable to expect, so check the project rather than trusting a comparison table, this one included. It is free to use, with a paid membership tier for the hosted network and a claimed name.

AppFlowy is the most straightforwardly positioned of the four: what you move to when Notion is the right shape but you do not want the data in someone else's account. Documents with block editing, grid and board and calendar views over the same records, templates, and a workspace built for more than one person.

It is written in Rust with a Flutter front end, giving it one codebase across desktop and mobile and a data layer that is genuinely local rather than a browser cache in front of a server. It is licensed AGPL v3, and the server half, AppFlowy Cloud, is published and self-hostable, so a small team can run the whole stack on its own infrastructure. A hosted option with a free tier and paid plans funds the project.

AI features exist and are moving. AppFlowy has added assistant features and, at various points, support for pointing them at local models rather than a hosted API. This is the fastest-changing part of the product and exactly the sort of claim that ages badly in an article, so treat the current documentation as authoritative. What a local model can usefully do over your own notes is covered in local-first AI at work and AI-enhanced PKM.

What it is not. Like Anytype, AppFlowy does not store your work as Markdown files in a folder; documents live in its own local database. Linking and graph features are lighter than a dedicated PKM tool's, because the emphasis is on structured records and shared workspaces rather than a dense web of notes. If you want a personal Zettelkasten, this is not it. If your team runs on a Notion workspace and the objection is where the data sits, it is the closest open replacement.

Licences and platforms are stable facts. Storage and sync are where the real differences sit, and where the consequences show up later.

One row does more work than the others. Zettlr needs no export because there is nothing to convert: the working files are already the format. For the other three, everything you write passes through an application-specific store, and the export function is the only bridge back to a general-purpose one.