Why I'm About to Start Talking to a Necklace

The best idea I had last week showed up a mile into a walk. It arrived complete: the framing, the caveats, the first sentence of how I’d explain it to you. By the time I was back at a keyboard I still had the topic. The idea itself, the living part, was gone.

Sound familiar? How many good ideas have you lost on a drive? How many solutions dissolved somewhere between the shower and the towel? How many 2am insights survived only as a note that says “database thing??”

I finally got tired of it, and my fix is admittedly a little strange: in about a week, I’m going to start talking to a necklace.

The bottleneck was never thinking

I run a personal AI system. (I think of the whole discipline as a Life OS, borrowing shamelessly from Daniel Miessler’s PAI work.) One of its core practices is what Miessler calls articulating your ideal state: saying out loud, in plain words, what you want some part of your life or work to actually look like, then refining that articulation over months as your understanding sharpens.

The practice works. The capture doesn’t. My best articulations don’t happen at a desk; they happen on walks, in the car, in the shapeless ten minutes between things. An AI system can only remember what I actually get into it, and my transfer rate from brain to system is, honestly, terrible.

So the project is simple to state: put a microphone where the ideas are, and build a path from there into my system’s memory.

I thought I was picking a gadget

The AI wearable space is suddenly crowded, so I did what I do with every real decision now: I took it to Claude and we argued about it at length. Four contenders earned a serious look, and each one came with an asterisk:

• Pocket is a lovely little capture device that keeps full transcripts of your own voice behind a Pro subscription.
• Fieldy stores your recordings in their cloud, and only their cloud.
• Limitless is built for always-on lifelogging, with an API still in beta.
• Plaud has a strong hardware story and an official API you join a waitlist to use.

Notice what those asterisks have in common. None of them are complaints about hardware. Every one is a dependency: on a subscription tier staying generous, on an API leaving beta, on a waitlist opening, on a company’s cloud outliving the company.

Somewhere in the middle of that debate the real question finally surfaced, and it ended the comparison on the spot: five years from now, can somebody else’s product decision quietly break my pipeline?

For every device on that list, the answer was yes.

The one left standing

Omi’s answer is no — the firmware, the backend, and the app are open source, and the whole stack is self-hostable.

That’s the entire decision. There was no runner-up, no feature matrix with a narrow winner. Once “no vendor can break this” crystallized as the axis that mattered, exactly one device was still standing. If Omi-the-company loses interest tomorrow, I can run the stack myself. Nothing about my memory should depend on someone else’s roadmap.

Is Omi the slickest product of the bunch? I have no idea yet; the box hasn’t arrived. It didn’t win on polish. It won on the only axis that matters for something I intend to still be using long after this product cycle ends.

Buying the device is the easy half

Building things for a couple of decades has taught me an ugly truth about capture systems: capturing is the easy half. The graveyard of productivity tooling is full of beautifully engineered pipelines that nothing ever read back out of. So before the hardware even ships, I wrote the plan down, and the plan is mostly rules for protecting the project from me:

Raw is forever; everything derived is disposable. Every transcript gets kept, permanently, untouched. Everything my system derives from a transcript can be thrown away and regenerated, because my understanding of this practice will change, and when it does I want to re-process years of raw thought with the better understanding. Keep only the processed output and every improvement orphans your history.

The machine proposes; I decide. Nothing lands in my curated memory without me reading it first. An automated writer will eventually, quietly, sand the edges off something I said on purpose.

Habit before pipeline. The first five days with the device involve zero code. I wear it, I talk, and I find out whether the articulation habit actually forms. The pipeline will still be two days of work next month. The habit either forms in the first week or it doesn’t.

What happens next

The pendant lands at the end of the month. Phase one is five days of doing nothing but talking to a necklace, which my family is going to love.

After that comes the build, and I’ll write down what actually happens: what the transcripts really look like, whether the habit survives contact with real life, and how much of this plan dies in week one. Plans written before the hardware arrives are cheap.

That’s exactly why I wrote this one down. When reality starts disagreeing with it, I want a record of what I believed… and a raw transcript of the moment I changed my mind.