DEV Community

Xin Jiang
Xin Jiang

Posted on

How a Thought From a Walk Becomes a Zettelkasten Note

How a Thought From a Walk Becomes a Zettelkasten Note

Your best reasoning shows up on the walk home, and by the time you sit down at the vault, all that's left is a keyword. You know the inbox note full of fragments like "seq problem order" that you can no longer rebuild into an argument. Zettelkasten has a step for capture, elaborate, file and link, and every one of those steps assumes you're sitting at a keyboard. There's no step for the thought that turns up somewhere else.

This post describes one way to close that gap: talk through the reasoning while you're still walking, and have it arrive in your vault that evening as an ordinary Markdown fleeting note. You then do the Zettelkasten work on it yourself. The tool is SpeakPen and its Obsidian community plugin, SpeakPen Sync. Below is how the capture works, what lands in the vault, the rule that keeps your rewrite safe, and what the tool doesn't do.

Why keywords and voice memos both fail

Picture it: you're halfway home and a problem you've been stuck on for weeks finally splits in two. You can feel the sentences of a permanent note forming. You're on a sidewalk, so you type three words into your phone and keep walking. At the desk those three words give you back the topic but not the proof.

The usual alternatives fail in their own ways:

  • Keywords keep the topic and lose the reasoning. Reasoning is the only part a permanent note actually needs.
  • Voice memos keep the reasoning, but they stay in the recorder app. You can't skim, search or link a recording, so each one is really a promise to make a note later, and those promises pile up.
  • "I'll remember it" works for moods. It doesn't work for arguments.

All three have the same problem: the thought has to cross from the sidewalk into the vault, and nothing carries it across.

The capture step

On the walk: start a recording without unlocking into an app. On iOS you can use the Control Center, Lock Screen or Action Button control, or ask Siri to "Record with SpeakPen". On Android there's a Quick Settings tile, and on desktop you can use speakpen.app/app in a browser. Record the reasoning itself, not a conclusion. It's fine if it's rough, with restarts and traffic noise. Thirty or forty seconds of thinking out loud holds more of the argument than three typed words.

What happens next: SpeakPen transcribes the recording and detects the language automatically. It then writes a note with a title, a summary, a category and the full transcript.

In the vault: the next time the plugin syncs, that note shows up as a plain .md file with YAML frontmatter, in the folder you picked. It isn't a special note type, so your links, queries, templates and Dataview views treat it like any other note. The title and summary are usually enough to tell from the file list whether a note is worth opening tonight.

At this point you have a fleeting note that still contains the argument.

The Zettelkasten step is still yours

The tool hands you back what you said, and you do the thinking from there.

  1. Read the note against your own memory of the walk. The transcript usually holds more than you remember saying, and that's why it's useful.
  2. Pull out the atomic idea and rewrite it in your own words. Keep the sentences that are already right, cut what's padding, and add whatever you notice while reading. That extra part is often the real insight.
  3. Split it if needed. A ninety-second recording can hold two ideas. You can see the seam between them on the page, which you can't do from memory, so make two permanent notes.
  4. Link each permanent note into the concept notes where it belongs.

The summary is not your permanent note. If you link it unchanged, you're filing someone else's wording as if it were yours. The value is that you get the full argument back to argue with, rather than a keyword you have to reconstruct from.

The rule that makes rewriting safe

This is the mechanism that matters most for this workflow. You edit synced notes in place, so a later sync must never overwrite your rewrite.

Here's how the plugin handles it. When it writes a note, it records a hash of the file's content. On later syncs, if SpeakPen has a newer version of that note (for example, you re-transcribed it), the plugin first compares the file on disk with that hash. If they differ, you've written in that file, so the plugin skips the file and tells you which notes it left alone. Once you've edited a note, it stops receiving updates from SpeakPen. The original transcript is still in SpeakPen if you need it.

Practically, this means you can rewrite the synced fleeting note directly, without copying it somewhere else first, and the plugin won't touch it again.

What it doesn't do

Readers here are right to check what a tool can do to their vault, so here are the limits:

  • Sync goes one way: from SpeakPen into the vault. The plugin makes read-only GET requests to speakpen.app and nothing else. It doesn't upload your vault and doesn't write anything back to SpeakPen. If you edit a note in Obsidian, that edit stays in Obsidian.
  • Syncs are incremental. Each sync asks only for notes that changed since the last one, using a cursor. A routine sync doesn't re-download your whole history or scan the vault.
  • Audio isn't linked in the note, on purpose. The API gives out presigned audio URLs that expire within hours. A link that's dead by next week would be worse than no link in a permanent note. Each note carries a stable speakpen_id in the frontmatter, which you can use to find the recording in SpeakPen.
  • It doesn't do Zettelkasten for you. It won't split ideas, choose links or write permanent notes. It only gets the reasoning across from the sidewalk to the desk.
  • It's not unlimited. The free plan is 3 notes a month, up to 5 minutes each. Pro is $9.90/month for 60 notes a month, up to 20 minutes each.

One optional extra: the same notes can also be read by ChatGPT, Claude or a coding agent through SpeakPen's read-only MCP server. It has three tools (recent notes, search and read one note), it never returns audio, and you can disconnect it in Settings → Connections. You might use it to ask "what did I say about sequencing on last week's walks?" before you sit down to write. Setup is at speakpen.app/connect. You don't need it for anything in this workflow.

The three-minute test

Setup takes a few minutes, and the free plan covers one honest experiment:

  1. In Obsidian, install SpeakPen Sync from the community plugin browser.
  2. In SpeakPen, go to Settings → Connections, find the Obsidian card, and click Generate token. The token is shown only once, so paste it into the plugin settings straight away and pick a sync folder.
  3. On your next walk, record about 30 seconds of reasoning, not keywords.
  4. That evening, open the synced note, rewrite it as one permanent note in your own words, and link it into your network.

The test is simple: did the reasoning make it from the walk to the vault? If you can still rebuild the argument from the note, it worked. If not, you've used one of three free recordings to find that out.

→ iPhone voice memos to Obsidian, end to end

Top comments (1)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.