I’m Using Claude to Manage My Health (And Track My Sodium)

I’m a web infrastructure guy. I spend my days thinking in systems — deployment pipelines, caching layers, failure modes. So when my January labs came back with a couple of flags, the natural move was to build a system around managing them.

The two things worth paying attention to: my Lp(a) came back at 153 nmol/L (anything above 125 is considered a cardiovascular risk-enhancing factor by the AHA), and my blood pressure has been consistently in Stage 1 hypertension territory. The Lp(a) is almost entirely genetic — there’s not much I can do about it directly. The blood pressure is a different story. That one I can actually work on.

The Problem With “Just Google It”

I could read articles. I could join Reddit threads. I could ask my doctor to explain everything at a 15-minute annual visit. None of those actually work the way I need them to.

What I needed was something that could hold all of my context — my actual lab values, my medication list, my history — and help me reason through it. Not give me generic wellness advice, but actually engage with my specific numbers and what they mean together.

So I set up a dedicated Claude project, uploaded all my lab PDFs, and started treating it like a knowledgeable health co-pilot. Not a replacement for my physician. A way to show up to those appointments better prepared, with the right questions already formed.

It’s been genuinely useful. When I asked about my Lp(a), it didn’t just tell me “that’s elevated.” It walked me through why it matters specifically in the context of my other markers — my ApoB is actually low-normal at 70 mg/dL, my LDL dropped from 122 to 96 between my April 2024 and January 2026 panels, my CRP is 0.7 (low inflammatory risk). So the picture is: one significant genetic risk factor sitting on top of an otherwise favorable lipid profile. That context changes how you think about it.

Blood Pressure Is the Lever I Can Pull

Lp(a) is a spectator sport. Blood pressure is not.

Stage 1 hypertension is systolic 130–139 or diastolic 80–89. My office readings have been bouncing around that range. The lifestyle interventions that actually move the needle on systolic BP — with meaningful evidence behind them — are pretty clear: aerobic exercise and sodium reduction. I’ve restarted my morning runs. The sodium piece needed more structure.

Here’s the thing about sodium: it’s not the salt shaker. Most people who think they eat “low sodium” are underestimating by a lot, because the sodium is buried in restaurant food, deli meat, bread, condiments, canned anything. You can’t manage what you’re not measuring.

Texting Claude My Meals

Starting now, I’m using Claude to maintain a running daily sodium tally. The workflow is simple: I send it a message describing what I ate, and it approximates the sodium content based on standard values for that type of food. At the end of the day I have a rough running total.

It won’t be perfect. Estimating sodium from meal descriptions is inherently approximate — a “turkey sandwich” can range from 600mg to 1,800mg depending on where it came from and how it was made. But directional accuracy is enough. The goal isn’t lab-grade precision. It’s building awareness of where the sodium is actually coming from and whether my daily total is trending in the right direction.

The target for pre-hypertension prevention is generally under 1,500–2,300mg/day, depending on which guideline you’re looking at. I’m starting without a firm number and letting the data tell me where I’m actually landing before I set a hard ceiling.

Why Claude Specifically

I’ve tried food logging apps. The friction of looking up every item, finding the right entry in a database, logging serving sizes — it doesn’t stick. Sending a message the way you’d text a friend takes about ten seconds.

What makes Claude useful here versus a notes app or a spreadsheet is that it can handle ambiguity. “Had a bowl of pho at lunch” — it can work with that. It knows roughly what’s in pho, can flag that restaurant pho tends to run high in sodium, and can ask a clarifying question if needed. It’s also holding all my health context, so it can connect dots over time rather than treating each meal in isolation.

How It Actually Works (The System)

The tracking setup is more structured than it sounds. I built a dedicated health folder in my Claude Code workspace — ~/Claude/health/ — with a CLAUDE.md instruction file, a comprehensive health context document containing all my lab values, medications, and clinical notes, and structured JSON files for sodium logs, BP readings, and exercise sessions.

OpenClaw (my personal AI agent running on Telegram) acts as the interface. I send it a meal description or a BP reading over Telegram. It passes that to a Claude Code session running in the health directory, which estimates the sodium, logs it to the structured data file, and reports back the running daily total. The data lives on my machine in version-controlled files — not in some app’s cloud.

The unexpected part: I can just send a photo. I snap a picture of a nutrition label, a restaurant menu, or a plate of food and send it over Telegram. OpenClaw reads the image, estimates the sodium from what it sees, and logs it. No typing out ingredients. No looking anything up. Just a photo.

First entry: my standard breakfast — bulletproof coffee (decaf, unsalted butter, MCT oil, milk), creatine, collagen, and a small amount of egg with a bit of cheese and ham. Total: 335mg sodium. The ham was doing most of the work at around 300mg. A useful data point on day one.

The agent also noted it would pre-fill that breakfast pattern for future days since it’s a daily constant. That kind of memory across sessions — keeping track of what’s routine — is what separates a conversational AI workflow from a generic logging app.

This Is the Right Use Case for AI in Personal Health

Not diagnosis. Not replacing clinical judgment. Maintaining context and doing the approximation work that humans are bad at doing consistently. The friction has to be low enough that it actually becomes a habit — and for me, that means it has to work the same way I communicate with anyone else: text a message, send a photo, get a useful response.

I’ll report back on whether this actually moves my numbers.

---

Related Reading

• How I Use Claude Every Day — The daily AI workflow that made health tracking a natural extension
• Two Claudes, One Telegram — The Telegram setup that makes texting Claude for health logging possible