What Happens Between the Sidewalk and Your Vault

You have a folder of voice memos you're never going to open. Each one held a thought that was alive when you recorded it, and each one is now a duration and a waveform — no title, no first line, no way to know what's inside without listening to the whole thing. Recording was never the problem; everything that should happen after the recording is.

Say you're walking home and you talk through a problem for forty seconds — half-sentences, a correction, a car going past. "The meetings run long not because of the agenda, it's that nobody says who decides." That thought is worth keeping. In a voice memo app, it's worth nothing until someone listens to it again, and that someone is you, at a desk, later, when you'd rather not.

SpeakPen is built to do the part in the middle so you don't have to. This piece walks through exactly what it does between the moment you stop talking and the moment a Markdown file shows up in your vault — and where it stops.

Why "just record it" never worked

Between "you said something" and "there's a usable note in your vault" sit a handful of jobs, and with a plain voice recorder every one of them is yours:

1. Turn speech into text. Otherwise you can't skim, search, or quote it.
2. Name it. A title is what makes a list of notes scannable. "Recording 038" is a box, not a note.
3. Say what's in it. A summary is what lets you decide whether to open it without replaying it.
4. Give it a handle. A category, so it's findable and queryable later.
5. Put it where your notes live. As an ordinary Markdown file in your vault — not a row inside someone else's app with an export button.

Done by hand, that's easily twenty minutes per recording, on top of listening to it at full length. Nobody pays that every week, so the recordings pile up and die. It isn't a discipline problem; the cost per thought is simply higher than the thought feels worth at 9 p.m.

What SpeakPen does after you stop talking

Capture. You record from wherever is fastest: on iPhone, the Control Center, Lock Screen, or Action Button control, or Siri ("Record with SpeakPen"); on Android, a Quick Settings tile; or the web app at speakpen.app/app. Language is detected automatically, so you don't pick one before you speak.

Processing, on the server. The audio is transcribed, then an AI pass writes a title, a short summary, and a category for it. The full transcript is kept as-is, false starts included — the summary is the readable layer, the transcript is the record of what you actually said.

Delivery, by the Obsidian plugin. The community plugin SpeakPen Sync checks in on an interval (every five minutes by default, or on demand). It asks the SpeakPen API only for notes that changed since its last sync, then writes each finished recording into the folder you chose as a plain Markdown file:

---
speakpen_id: "…"
title: "Meetings run long because nobody owns the decision"
category: "…"
created_at: …
synced_at: …
---

## Summary
…

## Transcript
…

The file is named after the title plus the date, so the sync folder reads like a list of thoughts rather than timestamps. From there it's a normal note: link it, tag it, query it with Dataview or Bases, move it.

Updates, without clobbering you. If a note changes on the SpeakPen side later and you haven't touched the copy in your vault, the plugin updates that file in place. If you have edited it, the plugin leaves it alone: it remembers what it wrote, compares before each sync, skips any note whose contents no longer match, and tells you which ones it skipped. Your words in the vault are never overwritten. That's the rule the whole design hangs on — the transcript still lives in SpeakPen, but your edits exist only in your vault, so your vault wins.

What it doesn't do

This audience checks, so here are the boundaries plainly:

• Sync is one-way, into the vault. The plugin reads from SpeakPen and writes files into your sync folder. It does not upload your vault and does not send your edits back to SpeakPen.
• It doesn't touch notes it didn't create. The rest of your vault is outside its scope.
• Audio is deliberately not linked from the note. The API hands out presigned audio URLs that expire within hours, so a link written into a permanent note would be dead by tomorrow. Every note carries a stable speakpen_id instead, which is how you find the recording again in SpeakPen.
• The summary is a first draft, not your thinking. Sometimes it's sharper than your memory of the thought; sometimes it flattens something that felt electric on the sidewalk. It's there so you can scan a list and decide what deserves your attention. Deciding what a thought means, and what it connects to, stays your job.
• It isn't a meeting recorder. It's built for short bursts of your own reasoning, not hour-long calls.

The same notes, readable by your AI tools

Once the notes exist, there's a second way to reach them. SpeakPen runs a read-only MCP server (setup at speakpen.app/connect) that ChatGPT, Claude, or a coding agent can connect to. It has three tools — list recent notes (up to 50), search (up to 10 results), and read one note — and nothing that creates, edits, or deletes. It doesn't return audio, it's included on every plan, and you can disconnect it from Settings → Connections. Useful when the question is "what did I say about onboarding last month?" and you're not at the vault.

What the evening looks like instead

The old ritual: open the memo app, play recording after recording to find out which ones mattered, then type up the survivors. Nineteen recordings was a lost evening, so it stopped happening.

The new one: open the sync folder, read the titles and summaries, keep the three that still matter, link them where they belong, archive the rest. The listening-and-labelling work already happened while you were making dinner. What's left is the part only you can do — reading your own reasoning and arguing with it.

A side effect worth noticing: when capture stops carrying a hidden twenty-minute bill, you stop pre-filtering half-formed thoughts before you've even said them. You record more of the things you used to let dissolve.

The three-minute test

1. Install SpeakPen Sync from Obsidian's community plugins.
2. In SpeakPen, go to Settings → Connections, find the Obsidian card, and tap Generate token. It's shown once — paste it straight into the plugin settings and choose a sync folder.
3. Next time you're walking, talk through one thing you're actually stuck on. Don't tidy it.
4. That evening, open the sync folder. Read only the title and summary.

If the title and summary tell you, without opening the note, whether that thought is worth keeping, the pipeline is doing its job. If they read like "recording 038", it isn't, and you shouldn't pay for it.

The free plan is 3 notes a month, up to 5 minutes each — enough to run this test. If it earns a place in your week, Pro is $9.90/month for 60 notes, up to 20 minutes each. The plugin itself is free.

→ iPhone voice memos to Obsidian, end to end