the second brain I already had

I had 249 voice memos on my phone.

Most of them were untitled. Some went back to 2017. All the “I should write that down” moments I’d ever had, stored on my phone and then immediately lost.

Last weekend I decided to actually do something with them.

---

here’s what I found: the capture problem isn’t a capture problem. it’s a retrieval problem.

I had been blaming myself for nine years for not being more disciplined, more organized, more consistent about journaling or taking notes. but I was capturing things. constantly. I just couldn’t find any of it.

248 voice memos. 58.8 hours of audio. Ideas, half-thoughts, decisions I made in the car, brand thinking I did out loud, company history I narrated to myself at 11pm when the kids were asleep.

all of it was there. I just didn’t know what was in it.

---

what I built

I spent a weekend building a skill file that:

1. transcribes every voice memo locally on my mac using whisper.cpp (no API keys, nothing leaves the machine)
2. runs each transcript through a local model to generate a title, a 2-3 sentence summary, and a list of themes
3. builds a master index file: one entry per memo, organized by theme
4. flags anything that looks like a distinct content atom (a story, an insight, a decision with a named outcome)

the whole thing runs offline. the cost was time to set it up, not dollars. once it was running I let it go overnight.

when I woke up, I had a searchable index of nine years of my own thinking.

---

what was in there

68 of those memos turned into distinct atoms in my content library.

including: the actual origin story of Helm. I remembered recording it but I didn’t know where it was. it’s a voice memo from July 2025, a minute and forty seconds long, where I’m describing the moment I realized I needed a focus tool for ADHD brains and why no one had built it right yet. it was sitting on my phone the entire time.

there was also an entire brand narrative I recorded in early 2025 and completely forgot. it’s basically the thesis for what Little Might is. I recorded it in a parking lot apparently.

there were memos from 2020 when I was in the middle of the partnership situation. I have no plans to turn those into content, but they matter as personal history. they’re in the index now. I can find them.

there were ideas I had for things that I’ve since built, and ideas I had for things I still haven’t. you get a weird feeling reading your own thinking from three years ago.

---

why this matters for AI-native building

the usual framing for “second brain” tools is: capture everything → organize it → retrieve it when you need it.

the part people skip is that the retrieval step was always the broken one. your brain captures. your phone captures. your notes app captures. none of them surface things when you actually need them.

what changed in the last year isn’t that AI got better at storing things. it’s that AI got good enough at summarizing and indexing things that you can finally make use of what you already have.

I didn’t need a new capture system. I needed to process what was already there.

and because I can run whisper.cpp locally on my mac, the privacy question isn’t a question. none of those parking lot voice memos are on a server somewhere. they’re transcribed locally, indexed locally, and I decide what I do with them.

---

the ADHD thing

I’ll say this once and not make it a whole section: if you have an ADHD brain, the “why didn’t I write that down” shame is usually misdiagnosed.

you did write it down. you recorded it. you took the screenshot. you sent yourself the email. the problem is that your systems aren’t optimized for retrieval, they’re optimized for capture, and ADHD brains capture in bursts.

what I built this weekend is a retrieval layer for the capturing I was already doing.

it won’t work for everyone. but if you’re sitting on years of untitled voice memos too, this is worth a weekend.

---

the setup, for those who want it

the short version:

• install whisper.cpp locally (homebrew: brew install whisper-cpp)
• grab your voice memos from the iPhone voice memos folder (they sync to iCloud, just find them in Finder)
• run whisper on each file to generate transcripts
• pipe the transcripts into Claude (local or API, your call) with a prompt like: “give me a title, a 2-sentence summary, and 3-5 themes for this audio transcript. flag if it contains a distinct story or insight worth developing.”
• build an index from the outputs

the skill file version I built handles all of this in one run. I’ll share the full thing in a future issue once it’s cleaned up enough that other people can use it without reading my internal OpenClaw docs.

---

the actual takeaway

the question worth asking isn’t “what should I capture?”

it’s “what have I already captured that I can’t find?”

nine years of my own thinking was sitting on my phone. I had no idea what was in it. now I do. that changes what I write, what I remember about where ideas came from, and honestly what I think about how my brain works.

that’s the whole thing.

---

if you’re sitting on a backlog of untitled voice memos, reply and tell me how many. I’m genuinely curious if I’m the only one.

---