Sisyphus,
Flipped

Circadian entrainment, audio journaling, and AI as a cognitive prosthetic.

I left my first rehab two weeks early. The official excuse was a bad roommate situation. The real reason was $8,000 for egregiously suboptimal therapeutic care. In places that confiscate your phone, the least they can do is give you a real replacement — not for scrolling, but for remembering what time actually feels like.

The second time I pulled up to the first rehab I went to, I left two weeks early.

I gave them a ▮▮▮▮ excuse. The real reason was that I was paying $8,000 for suboptimal — egregiously suboptimal — therapeutic services, and I didn't know how to tell them that without getting into a whole ego debate with the facility. So I said the roommate and I had deep history and beef, and I wasn't sure he wasn't going to go postal. That part wasn't entirely a lie. I was genuinely uncertain.

But if the facility had at least provided supplementative therapeutic care, I think I would have stayed the whole way through.

I don't think even that would have magically fixed my specific case. What I do think is that in rehabs where they typically take your phone, it is most necessary to provide a replacement. And that replacement is not for texting other people. It's not for entertainment.

It's for circadian
entrainment.

What Circadian Entrainment Actually Means Here

Circadian entrainment is the process of reminding the brain how long time actually occurs within — how fast time moves, what a real hour feels like when you're cut off from normal cues.

When your phone is taken, you lose one of the most powerful external clocks most people have. Without it, days blur. Therapy that is already suboptimal becomes even less effective because the basic rhythm of the day is broken.

A shallow fix would be a basic alarm engine. We can do much better.

With user permission, through a defined microphone source, the system can capture audio clips in three varying sizes — structured, intentional audio journal entries. Not voice memos for later entertainment: short enough to be sustainable, long enough to capture real reflection.

The Hardware Trap (And How to Avoid It)

The tricky part is bypassing the need for a custom manufactured device. Most "solutions" don't account for the fact that the human body moves. You're going to be far from the microphone at some point. I contemplated a necklace, a wristband, or some little companion device that pairs with the phone so you could raise it to your face and speak into it.

But that would make the entire problem space ten times harder. Now you have to design PCBs, understand manufacturing tolerances, deal with ordering runs, and pray you didn't get anything a hair wrong on the first revision. That path turns a simple app into a hardware company.

The beauty of circadian entrainment is that you don't need the necklace. You can use the phone you already have — or the replacement device the facility provides — and build the entire interface in software.

Sesefus — Sisyphus, Spelt Differently

This is the core idea behind Sesefus — the CLI command is ssfs.

Sisyphus is about flipping the folk tale on its head. Instead of being punished by endlessly pushing a boulder up a hill — or, in the modern version, burning out from your own intellect — you utilize AI as a cognitive prosthetic. The goal is to better your own mind in a way that doesn't require money, healthcare insurance, or a professional — though it would benefit greatly from a professional's assistant. Because of the irony, we spell it differently: Sesefus.

The backbone is the command-line interface. All actions the app can take for the user are referenced by the brief ssfs prefix, followed by category, command, variable flags, and other variables.

Rich Alarm Interface on a Shared Architecture

The alarm system is not just "set a reminder." You'd be able to:

• ▸Group alarms by day.
• ▸Further group them by category of intent.
• ▸Save and swap entire profiles for the week.

Example: you want all your Mondays at 9am to trigger a particular thing. Save that configuration as a profile. Later you can swap the entire 9am slot between different profiles without editing one by one by one.

Future machine-learning features include using embedding spaces to match verses of religious texts to times of the day, localized to specific journal prompts. If a prompt scores highly against a verse from the Bible — or the Quran, or whichever text you affiliate with — the system can surface a curated notion at the next alarm that resonates strongly with what you're working on.

Audio as First-Class Data

The three clip lengths feed into an audio database. From each entry, a 1-to-1 artifact database is derived — transcriptions, embeddings, metadata. This processing happens offline on the host, not in real time.

The architecture is deliberately a host + embedded client model. Run the full host+client bundle on one machine, or copy the client executable across devices on your local network for a heterogeneous compute stack. The client handles local audio capture (Win32 waveIn or equivalent); the host manages schedules, vault storage, and coordination.

This is already partially implemented in Zig as the Circadia / Sesephus alarm engine — host daemon plus edge clients communicating over length-prefixed TCP frames, with an HTTP control surface and audio journaling paths.

Local AI, Not Cloud Dependency

Sesefus is designed to scale to localized, hardware-accelerated machine-learning inference on the audio and artifact databases.

The vLLM porting work happening in parallel (Crow-9B-HERETIC via a local OpenAI-compatible endpoint) is part of the same prosthetic layer — bringing high-quality language-model assistance for prompt matching and reflection suggestions without sending your private reflections anywhere.

Why This Matters

In recovery environments — and honestly, in ordinary life under stress — the difference between "I have a phone taken away" and "I have a structured, private, time-aware prosthetic that helps me remember what a day actually is" is enormous. It turns the confiscated phone from a punishment into an opportunity for a better-designed replacement.

No custom hardware. No $8,000 for half-baked therapy. Just reliable software, audio that respects the user's real state, rich scheduling that actually matches how humans live their weeks, and local AI that acts as an assistant rather than a replacement for a real person.

Sisyphus doesn't have to push
the boulder alone anymore.