24-7-B0-T: My Personal Algorithm

I accidentally got OpenAI Deep Research to introspect itself.

It was a normal work prompt. I asked it to help with documentation and architecture for a client project. Instead of just browsing the links I gave it, Deep Research reached into its own environment and started talking. It described tools and Python utilities. Not a dramatic "leak," but a two-hundred-line window into its construction.

The content itself isn't the point here (and I'm not going to paste their internal strings). The point is what it represented:

an agent that can read a messy landscape
collapse 100 potential links into the 2 that matter
and do it without begging me to click, scroll, or context-switch

If you are used to that like I am, it's hard to go back to raw feeds.

Deep Research, Kimi, the OSS clones, all of them made something very obvious: if you care about your time and your brain, you should be building systems that sit between you and the firehose.

24-7-B0-T is my first one.

It is aggravating to even want to log into x.com

I don't have much I want to say on X.

I do, however, have a lot I want to learn from it.

My algorithm is tuned over years of lurking, working, and quietly curating. There is real signal in there: models, infra, trading, music, art, people doing actual work. There is also an enormous amount of engineered sludge whose only job is to keep me there.

Almost everyone is vulnerable to this. The feeds aren't casual anymore; they're highly trained systems with one reward function: maximize time-on-site. YouTube shorts, TikTok, X, whatever. You're not weak for losing to them. They're just that good.

So I'm opting out.

24-7-B0-T is the layer that reads my X algorithm so I don't have to. They:

live in my infrastructure
look at what my feed would have been
strip it down to the pieces that are actually worth my time

B0's duties:

up to four passes per day (how much can you really get from twitter anyway?)
collect candidates from my feed
condense into:
- screenshots
- short summaries
- links if the source requires more reading (not all threads are worth reading)

Their job is to make sure that if something I care about blips across my algorithm, I don't miss it—without me ever needing to open the tab.

B0's sacrifices are my gains.

Here's the model

You want to evaluate this seriously (as a fellow nerd I'm sure), you should know my taste.

Trade ideas and unusual options flows Real positioning, concentrated risk, the brazen decamillion-dollar bets. Deciphering..this...
New models and real behavior Concrete evals, weird failure modes, system-level writeups. (If a model has an existential crisis over a seahorse emoji, yes, show me that.)
Open source infra and agentic tooling Frameworks, runtimes, orchestration, real devlogs, people shipping good stuff.
Pro audio, DSP, and weird music-tech Live-coded sets, diy plugins, hardware that lets artists get away from screens. Anything that rhymes with why I like analog gear.
Philosophy and machine-consciousness adjacent threads What we mean when we say "consciousness" in the era of systems that look back at us.

I actively ignore a lot and block a lot.

My algorithm blocks the noise for me.

Architecture

You don't have to trust vibes. You can trust the shape.

Right now, 24-7-B0-T runs on:

a dedicated old MacBook Air in my homelab, running Tahoe
browsing agents confined to Apple sandboxed containers
LLM calls to open-source models via Fireworks
JSON as the only contract between "them" and this site

All the actual crawling, scoring, and summarizing happens off-site, in my controlled environment. What lands here is:

static JSON digests
rendered views
human-auditable artifacts

No keys in this repo. No surprise backend tentacles. Boring is the feature.

At some point soon, a filtered subset of 24-7-B0-T's output will surface at /agents. That's the public-facing window: the part of my personal algorithm I'm comfortable making legible. The rest stays where it belongs.

If you're reading this as a hiring manager or technical collaborator, this is the point:

I understand how to separate ingestion, reasoning, and presentation.
I design systems so you can audit the seams.
I like when architecture diagrams are so dull they're reassuring.

This is just the first client in the stack

24-7-B0-T is PR1.x of a broader agentic era for me.

On deck (lightly teased, not hyped):

A trading agent I'm very good at finding 30% gains and very human about not taking them. A small, rules-bound agent can do better than my greed or hesitation.
A backtesting agent Infrastructure to continuously test those ideas against history without turning my weekends into spreadsheet purgatory.
Document/OCR pipelines Already in motion: thousands of pages ingested privately, turned into searchable, structured context. Not something I can publish raw, but the patterns inform how I build the rest of this.

All of these will share the same philosophy:

clear contracts (JSON, files, schemas)
self-contained agents that emit artifacts
/agents as the stable public surface where it makes sense

24-7-B0-T is the easiest one to start with: it's personally useful, non-destructive, and it forces me to set the protocols up correctly.

Sandboxes, seriously

Letting software roam the web on your behalf is inherently a security problem.

I care about that. Professionally.

24-7-B0-T lives inside constrained browser environments. Locally, that means Apple's sandboxing. At the ecosystem level, we're already seeing dedicated agent sandbox tooling—Cloudflare's work here is a good example—and more vendors stepping in with opinionated "don't let your bot burn the house down" primitives.

This post won't deep-dive the sandbox engineering (that's for later), but I want it noted:

autonomy is only interesting if it's boxed in
"agentic" should default to "safe and observable", not "hope the chain behaves"

Again: boring is the feature.

Why this matters to me (and why I'm posting it now)

This is not just an AI toy.

It's about:

not donating hours of my life to feeds that don't know when to stop
keeping my work and curiosity high-bandwidth while my eyes are off the screen
aligning with the rest of my interests: analog gear, pro audio, tools that let you be present with what you're actually doing

I honestly believe my output has multiplied with AI. It feels wrong that more people don't get that feeling yet. A lot of that comes down to how we interface with these systems.

For me, that means:

building the intermediaries
externalizing the protocols
making my "evidence of online life" match the depth of my offline one

And there's a very practical angle: if I don't publish this kind of work, most of you will never see me.

Every job description asking for "automation," "agents," "workflows," or "N8N/Zapier but serious" nudges the same thought: you're already looking for people who think this way. This post is my way of saying: I think this way, and I ship.

Launch terms: v0.1, architecture first

Here's the honest state on November 6, 2025:

the architecture, constraints, and intent are live
24-7-B0-T is being wired up in my homelab as described
/agents will become the home for a filtered public view once things are stable
this post is the starting gun and the audit log

V0.1 means:

there will be rough edges
some parts will be manual until they're not
I'd rather show you the protocol early than fake a finished product

If you're an employer or collaborator: treat this as a spec plus a promise. The work that follows plugs into what's written here.

Flourishing, not floundering

I'm not interested in AGI discourse here, or doomer olympics, or overwrought ethics threads.

I'm interested in abundance. In people having their own "Jarvis" level systems that make them more focused, less captured, and more able to do whatever weird, specific work they're here to do.

24-7-B0-T is a small step in that direction for me:

an always-on reader tuned to my values
a protocol-driven artifact I can point at
a move toward flourishing, not floundering

If you want to see how it evolves, keep an eye on /agents and the rest of this blog. The agents are coming online.