Obsidian for Novel Writing: What Works, What Breaks

Obsidian is the tool most novelists land on after they get tired of their world living in twelve places. It's free, it's fast, your files are plain markdown on your own disk, and [[wiki-links]] are the right primitive for a story bible. A lot of very good books are being written in it.

It's also a note-taking application that has been talked into being a novel-writing application, and the seams show in specific, predictable places. This is a guide to both halves: a setup that works, and an honest account of what you'll hit.

We build a different tool, which we mention once at the end. Everything before that is written to be useful if you never leave Obsidian — which, if this setup works for you, is a perfectly good outcome.

A setup that works

Structure the vault around the story, not the notes

A vault layout that survives a full draft:

/Manuscript      chapters, in reading order, numbered
/Codex           characters, places, factions, items
/Notes           research, half-ideas, cut material
/Meta            outlines, revision plans, submission tracking

The important discipline is that /Manuscript contains only prose. The moment reference material starts living among the chapters, the word count is a lie and compiling gets painful.

Number chapters with leading zeros (01-, 02-) so they sort correctly forever, and give them names rather than only numbers — 12-The Drowned Chapel is findable, Chapter 12 is not.

Use frontmatter as structured data from day one

Every codex entry gets YAML frontmatter:

---
type: character
status: alive
faction: House Vale
first_appears: 03
---

This is dull to set up and pays for itself the first time you ask a question of your own book. With the Dataview plugin, that frontmatter becomes queryable:

```dataview
TABLE faction, status FROM "Codex"
WHERE type = "character" AND status = "alive"
```

A living cast list you never maintain by hand.

Link relationships instead of describing them

Write Sister of [[Corin Vale]], not Sister of Corin. The prose is nearly identical; the difference is that the first creates a connection Obsidian can see and the second creates a string only you can see. Backlinks then give you, for free, a per-character list of everywhere they're implicated.

Do this consistently in codex entries and also in chapter files — linking a character the first time they appear in each chapter turns the backlinks panel into a scene index.

Plugins worth the risk

Keep the list short; every plugin is a thing that can break on update.

• Longform — organizes chapters into a manuscript and compiles them to a single file. This is the one that most directly fills the gap.
• Dataview — queries over frontmatter, as above.
• Templater — new-character and new-chapter templates so structure is automatic.
• A word count or writing-statistics plugin, if daily counts motivate you.

That's enough. Novelists who install thirty plugins tend to end up maintaining a tool rather than writing a book.

Snapshot before every structural edit

Obsidian's file recovery is per-file and shallow. Before a major revision, commit the vault to git, or copy the folder somewhere dated. The cost is thirty seconds; the alternative is a subplot you can't get back.

Where it breaks

Four places, in roughly the order you'll hit them.

1. The graph looks impressive and tells you nothing

The graph view is the feature that sells Obsidian and the one that disappoints novelists fastest. Every note is the same dot. A character, a chapter, a stray research clipping and a shopping list are visually identical, so at 200 notes you get a hairball — pretty, undifferentiated, and useless for the questions you actually have.

The questions a novelist wants to ask are typed: show me only characters and the factions they belong to; show me which locations appear in act two. Obsidian's filters work on paths and tags rather than on the shape of your world, so the closest you get is a set of saved filter strings you maintain by hand.

Mitigation: use folder-scoped filters aggressively (path:Codex -path:Notes), colour groups by folder, and treat the graph as an occasional audit tool rather than a map. Check it for orphans — entries nothing links to — because those are your abandoned threads.

2. Nothing checks continuity

This is the real gap, and it isn't Obsidian's fault — it's out of scope for a notes app. Nothing in the vault objects when chapter nineteen contradicts chapter four. Backlinks tell you where a name appears, never whether the claims agree. The traditional answer is a continuity pass at the end of the draft, which is when your memory of your own decisions is weakest.

Mitigation: a chapter-boundary audit. When you finish a chapter, open every codex entry it touched and update it then, while you still remember why. Five minutes per chapter beats five days at the end.

3. Long-form editing in a note editor

Obsidian's editor is genuinely pleasant for notes and mildly wrong for prose. Scenes want a different set of affordances than notes do: distraction-free drafting, comfortable measure, revision marks, a sense of the whole chapter's shape. Longform covers the organizational half; the typing experience is still a note editor.

Mitigation: a dedicated writing-mode workspace — sidebars collapsed, readable line length, a font you like — that you switch to for drafting.

4. Collaboration and mobile

If you work with a co-author, an editor, or a sensitivity reader, Obsidian has no real answer. Shared vaults through Sync are approximately "one person at a time, politely." The mobile app is capable but is a vault browser more than a drafting environment.

Mitigation: export to a shared document for the editing pass and accept that the round trip is manual.

When it's worth changing tools

Not always. If your world fits in a few dozen entries, if you write alone, and if you enjoy the setup, Obsidian plus Longform plus Dataview is a genuinely excellent free stack and you should keep it.

The case for changing gets stronger when the two structural gaps start costing you real time — when the graph can't answer the questions you have because everything is the same dot, and when contradictions are surfacing in feedback rather than in drafting.

Those two are what Thread exists to close: documents are typed, so the graph can show characters, places and chapters as different things and filter to the question you're asking; and an AI consistency check reads a chapter against everything you established earlier and flags the disagreements. It keeps the parts of Obsidian's model that are right — plain markdown, [[wiki-links]], offline-first, export whenever you want — because those parts are right, and there's an importer that takes a folder of markdown so a vault comes across intact.

What you'd give up, honestly: the plugin ecosystem, and the certainty of files sitting on your own disk in a folder you control. Those are real losses, and for some writers they settle it.

If you want to see the typed-graph version of the idea before deciding anything, there's a demo world you can open without an account. And if you stay in Obsidian, take the chapter-boundary audit with you — it's the highest-value habit in this whole article, and it costs nothing.