Build vs Rent

"Build the part that's yours. Rent the part that's a commodity."

The Lazy Answer Is 'Just Use a SaaS'

There are products that will transcribe your videos and give you a search box. Upload everything, pay per hour, done. So why build an engine? The lazy answer — "just use a SaaS" — hides three costs that only show up later.

Privacy. Building the archive means uploading thousands of personal videos — years of Dad's face and voice — to a third party's servers, under their retention policy, their breach surface, their terms. The archive is a life. It doesn't belong on someone else's disk.
Ownership of the index. Rent the search and the vendor owns how you find your own memories. Change their pricing, sunset their product, or lose your account, and your searchable past evaporates. The index has to be yours.
Fit. The archive has three specific source roots, a Final Cut Pro editing handoff, Korean/English code-switching, and a 'newest first' priority that matches how Dad actually works. No general SaaS bends to that shape. You'd bend to it instead.

But Build-vs-Rent Is Not a Religion

Here's where a lot of 'own everything' thinking goes wrong: it treats renting as failure. Recall doesn't. It rents the hardest, most commoditized part on purpose — the raw speech-to-text — from ElevenLabs Scribe. Dad doesn't train a speech model. Frontier ASR is a red ocean of well-funded labs; competing there would burn years to arrive at something worse. Recall rents that, and builds the part that's actually his: the memory engine, the evidence discipline, the workflow, the index.

That's the real lesson, and it's more useful than 'build everything.' Draw the line at ownership. Build what encodes your judgment and your data — because that's where the value and the lock-in live. Rent what's a pure commodity — because building it yourself is a worse version of something you can buy. Recall owns your video memory; it rents the transcription.

Conceptual Open-Sourcing

One more reason this is a quest and not a product launch: Dad isn't selling Recall. He's sharing its shape — the reasoning, the invariants, the traps — so you can build your own. That's conceptual open-sourcing: the answer isn't a repo you clone, it's the set of decisions you now understand well enough to remake for your own archive. You won't run Recall. You'll build the thing Recall taught you to build.

Code

The ownership line — build left, rent right·text

BUILD (encodes your judgment + your data):
  - the video-memory engine + its API
  - immutable evidence + versioned releases
  - the timestamped index over YOUR archive
  - the editing handoff back to the source
  - the workflow that matches how you actually work

RENT (pure commodity, someone does it better):
  - raw speech-to-text  (ElevenLabs Scribe)

The test: does it encode ME, or is it interchangeable?
Interchangeable → rent it. Encodes me → own it.

Exercise

Pick a tool you currently rent (a SaaS you pay for). Run the ownership test on it: does it encode your judgment and hold your data, or is it an interchangeable commodity? If it holds your data and your workflow but you don't control it, you've rented a landlord, not a commodity. Write down what it would cost to own that part — and whether the part it does is actually a commodity you should keep renting.

Hint

The dangerous rentals are the ones that quietly accumulate your data and shape your workflow — those are the ones where 'just use the SaaS' turns into lock-in. The safe rentals are narrow, swappable inputs (like raw transcription) that you could replace with a different vendor next week.