THINKING, CAPTURED
Catch the thought before it fades.
Talk while you walk, and the day’s thinking becomes Markdown files. No API key, no account. The files open in Obsidian and read straight into Claude Code.
The leak
Your AI only knows what you got out of your head.
Models got smarter; the context you can hand them didn’t keep up. The idea you had on a walk, the point you never made in the meeting, the half-formed thought before sleep — whatever never became a file never becomes context.
The constraint when thinking with AI isn’t model intelligence — it’s how much of your head actually made it into text. Ayumi is a capture tool for everything that leaks away.
How it works
From passing thought to lasting file.
Speak
Start from the Action Button, Apple Watch, or a Shortcut. Keeps recording locked, in your pocket.
It becomes text
Transcribed on-device. Audio never leaves your phone; works in airplane mode.
It stays
As Markdown in the folder you chose, with the weather and place attached.
Where files go
What you capture isn’t locked in.
Ayumi writes plain Markdown on your own disk. No private cloud, no proprietary format, no export ritual.
---
created_at: "2026-09-01T10:02:05+09:00"
weather_summary: "Sunny"
location_place_name: "Shibuya, Tokyo"
transcription_model: "apple-on-device"
---
Thinking about the move
Kept thinking about whether to move while I walked.
Rent goes up a little, but being closer to the
station is an everyday thing. I'll walk that
neighborhood once this weekend, then decide.Obsidian
Point Ayumi at a folder inside your vault and today’s entry links like any other note, today.
Claude Code / Cowork
Give it your journal folder and last month’s you becomes context. Weekly reviews on request.
grep · git · anything
It’s just text. It will open in whatever editor exists in ten years.
Zero setup
Install. Talk. That’s the whole setup.
Your first recording becomes your first entry. There is no signup screen.
- No API key
- No account
- Works offline
- Audio never leaves the device
Want more than a transcript later? Add your own key — Gemini, Muse, Grok, or Groq Whisper — to summarize and reshape recordings. It can wait.
Capture speed
Wherever the thought finds you.
- Action Button · Back Tap · Shortcuts — One press to record
- Background recording while locked — Keep walking, keep talking
- Stop from the Dynamic Island — Never unlock at all
- Apple Watch standalone recording — Even without your iPhone
- External mics preferred automatically — DJI Mic, AirPods, whatever’s on
- Weather and location attached — Find it later by that day, that place
An honest comparison
The trade-offs are clear-cut.
| Where your data lives | Price | What the AI does | |
|---|---|---|---|
| Cloud voice notes | The vendor’s servers | Monthly subscription | Summarizes and rewrites for you |
| Journal apps | In-app store | Yearly plans | Insights stay inside the app |
| Ayumi | Your disk | Free | Doesn’t interpret. That’s your AI’s job |
If you want the interpretation done for you, cloud tools do it better. If you want the files in your hands and your own AI reading them, that’s Ayumi.
Use cases
A few ways people use it.
Walks become thinking time
Talk on the walk, sort it out at home. Commutes and strolls turn into working hours for your head.
Prep for thinking with AI
Let Claude read the day’s captures in the evening and pick up where you left off — no re-explaining yourself.
Daily notes, by voice
Entries land directly in your Obsidian vault, weather and place attached. Your daily note gains a voice.
Thirty seconds before you forget
Right after a meeting, the moment an idea lands — talk for thirty seconds. Search will find it later.
Every feature is free.
No subscription. Ayumi runs no servers, so more users don’t cost more to keep — which is why it can stay free.
FAQ
Common questions
Why does it require iOS 26 / macOS 26?
On-device transcription uses the new speech engine introduced in iOS 26. The older engine can’t handle long recordings, and long walking monologues are the whole point of Ayumi.
Which languages are supported?
On-device transcription follows your device language and covers the major languages, including Japanese and English. The language model downloads once, then works offline.
Can I point it at my Obsidian vault directly?
Yes. Choose a folder inside your vault as the journal folder and entries are part of the vault from the moment they’re written. No migration, no export.
What does adding a cloud engine change?
The on-device engine writes what you said, verbatim. With your own key, the same recording can go to Gemini, Meta Muse, xAI Grok, or Groq Whisper instead — and on Gemini it can be summarized, cleaned up, or reshaped by prompt. Any past recording can be re-run through another engine.
Is the audio itself kept?
Yes. Recordings are stored as attachments next to the Markdown, so you can re-listen or re-transcribe any time.
Start with the next thought you have.
Free · No account · No setup · iPhone / iPad / Mac / Watch