Concept

Ayumi is built around a small set of opinions about how to think with AI today. This page is the “why” behind the app — the principles that shape every feature decision.

Local-first by design

Every entry you write in Ayumi is a plain Markdown file on your own disk (or in your iCloud Drive folder, if you choose). There is no proprietary database, no opaque sync layer, no account to lose access to.

The folder you point Ayumi at is yours. You can open it in Finder right now.
Each entry is a .md file with YAML frontmatter and a .assets/ folder for attachments.
Nothing leaves your device unless you actively send it somewhere — for example, by configuring a Gemini API key for transcription.

This is the foundation everything else rests on.

You own your journal

A journal is the most private thing most people write. Doubts you haven’t told anyone, notes on your health and relationships, plans that haven’t taken shape yet — words that were never meant to be handed to someone else.

Meanwhile, data breaches at cloud services are no longer rare, and AI agents are automating attacks, making breaches larger and more frequent. Storing your journal on someone else’s server means accepting that you don’t really know where your most private words are kept, or how.

Ayumi is designed so you never have to think about that. Your journal is yours, so it lives as first-class files in secure local storage. AI works on those files only when you decide it should. Keeping those two things separate is the design decision Ayumi cares about most.

Your journal stays as files you hold. Entries are Markdown files on your own device, or in an iCloud Drive folder you choose. Ayumi has no server that holds your data and no account to sign up for. You know where your journal is, and you’re the one who controls it.
You decide, each time, whether anything goes to AI. Transcription runs on-device by default, so the audio never leaves your device. A cloud AI is used only when you’ve set up your own API key and picked a profile that uses it. You can switch engines per recording — keep this one on-device, send that one to Gemini to reshape it — one decision at a time.
You choose which AI provider to trust. Ayumi connects directly to the provider you pick — Google, Meta, xAI, or Groq — with your own API key. Ayumi never relays or stores your data in between. API keys are stored in Keychain.

Keep AI processing entirely local, or hand it to a provider you trust only when you need to. Either way, you stay in control of what leaves your device.

Markdown is the substrate of the LLM era

Markdown is durable, portable, and the format LLMs read most fluently. Today’s frontier models, last year’s models, and the models that ship next year all consume Markdown as a first-class citizen. Tools and agents come and go; the text on disk outlives them.

Choosing Markdown isn’t just an aesthetic preference. It means the same files that organize your thinking become first-class context for any generative AI tool you want to use — without conversion, without export, without lock-in.

The missing input for personal intelligence

AI assistants are becoming personal. They can now read your email, your photos, and your search history to answer in the context of your own life. That is useful, but look at what those sources actually hold: records of what happened — what you received, what you photographed, what you looked up.

What they don’t hold is what you thought about any of it. Why you made a decision, the doubt you haven’t voiced, the idea from this morning’s walk. That never reaches an inbox, so connecting more apps won’t hand it to an AI.

The companies building AI are now working on a personal superintelligence. In his letter Personal Superintelligence, Meta’s Mark Zuckerberg describes it as AI that knows you deeply and helps you reach your goals. But the best model changes almost every year. Whether you can move to the newest one each time depends on where the context that knows you lives. Inside one company’s account, it ties you to that company’s AI.

Keep your personal context under your full control, so you can always switch to the newest superintelligence.

Ayumi is built to capture that missing layer, and to keep it as files you hold rather than data inside someone else’s account. Whichever AI you choose can then work from the real you — on your terms. You can even hand a copy to Gemini: see Use Your Journal with Gemini Personal Intelligence.

The real bottleneck is capture speed

Most note-taking advice optimizes the wrong thing. Folder structures, tag taxonomies, templates, daily-note schemes — these are organizational concerns. They matter, but they aren’t the binding constraint.

The binding constraint, when you want to think with AI, is how much of what’s actually in your head you can get into text before it evaporates. Half-formed ideas, the digression you had on a walk, the thing you were going to say in the meeting but didn’t — none of it can become AI context if it never reached a file.

If you can’t dump your thoughts quickly and completely, the AI is working against a sketch of you instead of the real thing.

This is the gap Ayumi is built to close.

How Ayumi removes capture friction

Every feature in Ayumi serves one question: can you get the thought out before it’s gone?

Quick Entry from a hardware trigger. Bind the Action Button, Back Tap, or any iOS Shortcut to a one-press capture flow. No app switching, no unlocking into a notes app — just thought to file. See iOS Shortcuts for ready-to-use patterns.
Background voice recording. Start a recording, lock your phone, and keep walking. Talk to yourself podcast-style for as long as you want. Live Activity controls let you stop from the Dynamic Island without unlocking.
AI transcription that fits the moment. Pipe recordings through Apple’s on-device transcription for verbatim memos, or through Gemini with your own prompt presets when you want the AI to reshape the monologue into something more structured. See Voice Recording.
Automatic context capture. Location and weather are recorded with every entry, so the lived context around a thought is preserved without effort.

The goal is simple: the resistance between having a thought and having that thought on disk as Markdown should be as close to zero as we can get it.

Your files, your toolchain

Ayumi is intentionally narrow. It’s a capture tool — it isn’t trying to be your knowledge graph, your AI agent, your editor, or your reviewer. Once your thoughts land on disk as Markdown, you are free to use whatever tools you like for the next loop:

Claude Cowork — point it at your Ayumi folder and let it run multi-step jobs across your notes: weekly distillation, theme extraction across months of entries, batch reorganization, scheduled reviews. Because Cowork reads and writes local files directly, it pairs naturally with a journal that already lives on your filesystem.
Obsidian — drop the same folder into a vault and you get backlinks, graph view, and the plugin ecosystem, with zero migration.
Cursor, Cline, Claude Code in the terminal — open your journal as a workspace and edit in conversation with the AI of your choice.
Plain text editors — VS Code, Vim, Notes.app, anything that opens .md. Always works.
Your own scripts — grep, rg, git log, a Python notebook. The folder is yours; treat it like any other directory of text.

Switch tools whenever you want. The files don’t move. That is what local-first plus plain Markdown buys you.