I set out to build another fitness app. What came out was the logbook of the AI era.

For about two years, the most consistent thing about my training was that I’d quit it. A bad stretch hit. Runaway anxiety, work that didn’t end. The gym went first. Yeah, I know. Then it stayed gone. By the time I looked up, two years had passed and I was carrying 15 extra kilos I’d never quite decided to put on.

A belly where the abs used to be. Out of breath climbing six flights. A face in the mirror I didn’t recognize. Enough. I made one promise that morning, and it still stands: I’m never going to be out of shape again. Not for a year, not for the rest of it.

That promise is why Verxion exists. But the thing I launched today is not the thing I set out to build, and the gap between the two is the actual story.

The momentum trap

Took me too long to admit this: I already knew what to do. I’ve been training for twenty years. I’ve worked with serious coaches (including a PhD in sports science) and with nutritionists. I didn’t lack a plan, a method, or information. I’d done the climb back into shape before.

So when I came back, I went looking. The App Store has them all: workout loggers, macro trackers, all-in-ones with the onboarding quiz and the streak badges. I tried a handful. Each one stopped a layer above where my training actually lives. Drop sets, rest-pause, different RiR targets per set, properly periodized mesocycles — none of them had room for it. I’d have to flatten what I actually do into something more convenient to use, and within a month I’d quietly stopped opening any.

With twenty years of training behind me, the only diagnosis I could reach was that I’d lost the edge. My fault, not the tool’s.

It took me longer than I’d like to flip that. What I’d actually been losing wasn’t discipline. It was momentum and consistency. And once you lose them, every day you don’t train makes the next one harder, until “starting again” feels impossible. The spiral quietly reinforces doing nothing.

I’d come back at a different stage of life. I’m not twenty anymore, and I was asking for a different thing. Experience had taught me the key is consistency — and consistency is just momentum, compounded. I wasn’t trying to get back in shape; I’d done that before. I was trying to never be out of shape again. Ever. That’s a longer, more demanding game, and I wanted a tool built for it: one that could keep momentum from breaking in the first place, hold the truth of what I do, and still be around in ten years.

What I built first

So I started building the app I wished existed. My angle wasn’t revolutionary, and I knew it: honest and more nuanced data. Most apps model a “set” as weight × reps and stop there. But real training isn’t that tidy. I built first-class support for the set types people actually run: drop sets (with individual drop stages), rest-pause, pyramids, supersets, giant sets, AMRAP. Plus the per-set details most apps don’t bother with, like RiR and cadence. Do a drop set, you shouldn’t have to flatten it into three fake “normal” sets that lie about what happened. The log should hold the truth of the session.

But honest data was only the floor. What I really wanted was a tool that would support me on the days I didn’t want to train (which are the days that matter most). One that offered viable swaps when I couldn’t, or wouldn’t, eat exactly what was planned, without breaking the plan. One that adapted the workout when I didn’t have everything I needed. The simple ask underneath: never let me break the momentum, even when my head fails or life gets in the way.

The data part was good. It was also, I slowly admitted to myself, incremental. A better-shaped database around the same tired interaction. I was still asking you to open an app and tap a form. I’d just made the form more accurate. Everything beyond the data, the parts that needed the tool to behave more like a coach than a notebook, was still out of reach.

The agents changed the question

The turn happened while I was building the assistant layer on top. I wired up an agent so you could describe a session in plain language instead of tapping, and watched it parse “worked up to a heavy triple on squat, then a back-off drop set” into correctly structured data, drop stages and all, without a single form.

That’s when the schema problem stopped looking like the actual problem. The apps weren’t broken. They were just from before. Opus 4.5 changed everything, and you start to notice it in odd places, including the moment you open a fitness app and realize you’re operating a CRUD. My workflows everywhere else had quietly reorganized themselves around talking to agents that operated whole systems on my behalf. The fitness app, by contrast, kept asking me to come tap a form.

What I actually wanted wasn’t more features, or even a better schema. It was a different shape of interaction. Fluid, conversational. A system you could talk to instead of operate, where your data was open to being questioned rather than just displayed. Concretely:

Logging a session by talking it out instead of tapping forms, including the parts that don’t fit a “weight × reps” schema.
Asking “what am I doing today” and getting the answer.
Swapping an ingredient mid-meal (“no rice, give me sweet potato”) and watching the day’s macros reconcile themselves.
Adapting the workout on the fly when something’s off, without breaking the program underneath.
Having a real conversation with your own data — “how’s my squat trending?”, “did I actually hit protein this week?” — instead of pulling up charts to read on your own.

My body kept making the gap obvious. You finish a heavy set, you’re breathing hard, your hands shaking from a max deadlift or the tail end of a rest-pause. The last thing that fits the moment is unlocking your phone to tap through a form. Nutrition is the same: a small data-entry chore three to five times a day, forever.

Don’t read any of this as a complaint about the apps that exist. Most of them are genuinely well-built. Thousands of people use them every day, and that’s earned. They just aren’t what I needed now. What I needed was a category that doesn’t really exist yet.

So I tried to throw the whole frame out and ask a different question:

What kind of technological companion would I have wanted at any point in those twenty years, if I weren’t limited to what a fitness app has always been? And can we actually build it now?

The answer was very different from the Verxion I’d been building. That’s the product I launched today.

A platform, not an app

Verxion is the logbook of the AI era. An agent-first fitness and nutrition platform that plugs into the AI clients you already use: Claude, ChatGPT, Cursor, Claude Code. There’s no app to download, no tab to keep alive. You talk to your assistant the way you already do, from wherever you want, and it operates a full training system underneath.

Mid-set, you describe what you just did however the words come out:

“Did 5 reps at 100kg, reached failure”

“5x100, RiR 0, 3rd pin on the bench”

“Holy shit, reached 100kg, did 5 reps and I left nothing in the tank. Love it”

Infinite voices, one set. Talk, type, type with typos and shaking hands. Your assistant figures it out. The system logs them all the same way: reps, load, RPE, volume, PRs, all updated. Food works the same: “had 200g of chicken and a cup of rice” reconciles against your macro targets, pulling from ~2.5M products.

Or skip the talking entirely. Hand it a coach’s PDF and have your whole block scheduled. Snap a photo of a food label and have it logged. Throw it a coach’s Excel, the kind with periodization, cadences, rest schemes, set types nested across mesocycles, and watch it ingest the whole thing in one shot. Whatever you can capture, your assistant can ingest.

And “operates” is the right word. There’s a real platform under the conversation: periodized programs and mesocycles that adapt as you progress, diets that adjust to adherence instead of locking you to a calendar, cardio, body and physique measurements, coaching relationships, plateau detection, monthly recaps that read like a coach wrote them. Your AI writes the story behind the numbers.

Today it exposes more than 350 distinct operations an agent can call, all against a single source of truth for your training. None of it requires you to learn a screen. As far as I know, that makes Verxion the first MCP-first fitness platform. API-first is the next step, so anything that can hit an endpoint will be able to read and write your training.

That choice is intentional. The frontier labs are already shipping state-of-the-art chat, voice, OCR, and vision to billions of people, and most of us pay for the subscriptions anyway. Better to be the platform underneath the apps you already use than to build a watered-down version of any of it. Agents are also a one-way paradigm shift. They’re going to spread into more contexts and more devices. By living at the protocol layer, Verxion is compatible with whatever ambient assistants ship next: glasses, watches, cars. Your training travels with the agents instead of getting stranded in an app you have to keep updating.

And it cuts both ways. Custom clients on top are part of the plan from v1. API-first means anyone (a power user, a coach, an indie dev) who wants their own interface or automation can hit the same operations the assistants do. The platform stays open.

Built for agents, the roadmap stops sounding like sci-fi and starts sounding like plumbing:

Log a set by voice through an agent on smart glasses. No phone, no hands.
Derive this week’s grocery list from the diet you’ve got programmed, and when you let it, place the order.

Neither of those needs a new app. They need a platform an agent can drive, which is the thing I pivoted to build.

Forget the MCP, the agents, the schema for a second. Underneath is the one outcome I’d been chasing for twenty years and never quite locked: consistency becomes easy, not a challenge. The friction that breaks momentum isn’t there. When your head fails or life gets in the way, the system stays with you. That’s the whole point.

I’ll be honest about the bet I’m not sure of: MCP is a young protocol. The clients are still maturing, the UX conventions aren’t settled, and I’m wagering that “your training lives inside your assistant” becomes normal rather than niche. I think the friction math makes that inevitable. But I’d be lying if I called it a sure thing. That’s the fun of building early.

Your training, your meals, your body. That’s some of the most personal data you’ll ever generate, and Verxion treats it accordingly: EU-hosted, GDPR-compliant, and encrypted by default from the very first commit.

Beta means we build v1 together

Verxion is in open beta, free, no credit card needed. But “beta” here isn’t a soft launch with a countdown. It’s the actual plan: I want to develop this with the people using it. I’m one person who trains one way; there are edges, workflows, and assumptions I will never see on my own. The beta is how we find them. When it’s been sanded down by real use and I’m genuinely confident in it, that’s v1.

Everyone who’s in during the beta gets a lifetime-locked price: a special early-supporter rate that holds for good, before paid plans land. It’s not a scarcity gimmick. It’s the honest trade. You use it while it’s rough and tell me what breaks, and you keep that price forever. So if any of this sounds like something you’d use, register now. That’s the window.

And if you’d rather just watch first: follow along. I’m building this fully in public from here. The decisions, the numbers, the things I get wrong, the edges the community surfaces. This product is going to be pure dogfooding: I use Verxion every single day, because the promise I made myself still stands and this is the tool I built to keep it. You’ll see exactly how it evolves, in real time, from the person betting his own training on it.

One more thing, because the timescale matters. I’m not a coach or an athlete. I’m someone normal who wants to stay in the best shape of his life until the day he’s done. No bullshit, no hacks, no shortcuts. My plan is to use Verxion for the rest of that life. That means showing up every day to make it the best it can be, and only stepping back if someone, somewhere, builds something notably better. This isn’t a hobby project I’m running to see where it goes. It’s the tool I’m going to need for the rest of my life, and I’m building it that way.

Try it at verxion.ai. Tell me where it annoys you — that’s literally the point.