SpeakribeDictation for Obsidian
Obsidian keeps your notes as local files on purpose. A dictation tool that uploads your voice to transcribe it would quietly undo that. Speakribe is the matching half: local app, local notes, local transcription.
Why it helps here
People choose Obsidian specifically to avoid a cloud, then reach for whatever dictation is convenient and hand the same thoughts to a transcription service. The privacy property is only as good as the weakest tool in the chain.
Local end to end
Your vault stays on disk and your voice stays on the machine. No part of the chain reaches a server.
Markdown comes out clean
You get punctuated prose that drops straight into a note; Obsidian's own syntax and shortcuts work on it as usual.
Fast capture
The point of a daily note is catching the thought before it goes. A shortcut and a sentence is about as low-friction as capture gets.
Things people say into Obsidian
- “Daily note. Spent the morning on the migration plan, still unsure about the rollback path.”
- “Idea: a weekly review template that pulls in anything tagged follow-up from the last seven days.”
- “Book notes: the argument is that constraints improve creative output, with three studies cited.”
- “Link this to the note on caching strategies, they're making the same point from different directions.”
Setting up for Obsidian
If you chose Obsidian, you chose local files over a sync service, and the setup here is mostly about not undoing that decision by accident somewhere else in the stack.
Check the one setting that can leave your Mac
Speakribe has an optional post-processing step that can send transcribed text to a language model you configure. It is off unless you turn it on. If local-only is the reason you use Obsidian, leave it off and the whole pipeline stays on your machine.
Use a binding Obsidian's command palette won't claim
Obsidian is heavily keyboard-driven and users rebind aggressively. A modifier-only shortcut avoids the palette and every plugin that has claimed a chord you have forgotten about.
Dictate into the editor, not the switcher
The quick switcher and the search pane are both text fields, so dictation works in them — which is occasionally useful and more often means half a note has gone into a search box. Click into the editor first.
Add your vault's proper nouns
Note titles you link to constantly, recurring tag names, the people who appear in your daily notes. Add them to the custom-words list so the words you type as links come back spelled the way your vault already spells them.
One dictation, start to finish
A daily-note capture, spoken in about twenty seconds while walking away from a meeting. The competing option here is not typing — it is not writing it down at all.
Talked to Dana about the pricing page. Her view is that the annual toggle is the thing holding conversion back, not the price itself, because people can't tell what they're committing to. She wants to test showing the monthly equivalent underneath the annual number. I said I'd look at whether we can do that without touching the checkout, which I think we can. Follow up next week.
Talked to Dana about the pricing page. Her view is that the annual toggle is the thing holding conversion back, not the price itself, because people can't tell what they're committing to. She wants to test showing the monthly equivalent underneath the annual number. I said I'd look at whether we can do that without touching the checkout, which I think we can. Follow up next week.
- Plain text, no markdown. There are no brackets around "Dana" and no tag on the last line — every link and tag in this note is something you add afterwards.
- "Dana" came out right because it is a common name. A vault full of unusual names is exactly the case the custom-words list exists for, and it is worth seeding before you rely on daily capture.
- Five sentences from one continuous take. Spoken capture produces short declarative sentences, which happens to be the register daily notes want anyway.
What it won't do in Obsidian
No markdown by voice
You cannot speak a wiki link, a tag, a heading or a code fence. Saying "open bracket bracket" types those words. Linking is a manual pass after the text exists, which for capture is usually the right order anyway.
It has no idea what a vault is
No note titles are known to it, nothing is indexed, and there is no plugin. That is precisely why it works identically in Obsidian, a terminal and a browser — and why it cannot autocomplete a link.
Local only if you leave it that way
Transcription is on-device with nothing optional about it. The post-processing feature is the one part that can make a network call, and it is opt-in — but it is worth knowing it is there rather than assuming.
Questions people actually ask
- Does anything leave my machine?
- No. Transcription runs on your Mac and Speakribe has no server or account, which keeps a local-first vault genuinely local.
- Can I dictate wiki links and tags?
- Speakribe writes prose rather than syntax. Add links and tags with Obsidian's own shortcuts once the text is down.
- Does it work in the editor and in the quick switcher?
- Any focused text field, so yes to both.
- Is there really nothing that goes online?
- Transcription is entirely on your Mac, with no account and no server. The one thing worth knowing about is an optional post-processing step that sends text to a model you configure yourself; it is off by default, and if you leave it off nothing about dictation touches the network.
- Can I speak wiki links and tags?
- No. Brackets, hashes and every other markdown character have to be typed. Dictation gets the prose down; linking it into your vault is a separate pass, which is how most people capture anyway.
Ready to stop typing?
One payment. Every future update included.