The DIY coach is real — build it.  The write-off isn't automatic. That takes a signature.
agenticHSA
the diy longevity stack · what it can't do

Your AI coach can read your bloodwork. It can't sign a thing.

There's a beautiful build going around: wire your ring, your labs and your food log into an AI, and fire your longevity coach. It's right — you should build it. But a coach reading data you already own is orientation, not medicine. It can recommend the CGM. It can't make it pre-tax. That step needs a licensed human's name on it.

See what you can shelter → The part they can't build
no w-2 needed · you already did the hard part · the signature is the part we are building for
first, the honest part

The DIY stack works. Keep it.

Connect the sources you already own to an AI agent and ask for a weekly read. For sleep, HRV, macros and trend-spotting, this genuinely replaces the expensive part of a concierge clinic — reading data back to you. We'd never tell you not to build it. We build things like it too.

Ring / watch→ Food log→ Bloodwork PDF→ your AI agent→ "what should I fix?"
This is a real coach for orientation — and as of 2026 it's practically built in. Anthropic now ships Claude for Healthcare, with official one-click connectors for your labs (Function), your medical records (HealthEx), and your Apple Health. The reading isn't just cheap anymore — it's becoming a platform feature. Which is exactly the point: when reading-your-data goes cheap and native, the value was never in the reading. It's in the one step no platform will touch. Read on.
the wall · three things an AI coach structurally cannot do

Then you ask it: "can I write this off?"

Your coach says you'd benefit from the CGM, the standing desk, the therapy, the sauna. Good advice. Now try to make any of it pre-tax. Here's where the DIY stack stops — and it isn't a bug you can patch.

wall 01

It can't sign.

An AI recommendation is an opinion. A Letter of Medical Necessity is a legal document under IRS §213(d) — and only a licensed physician can write one. No prompt turns "you'd benefit from this" into eligible spend. The model can draft the letter; it cannot be the signature.

wall 02

Nothing has a name on it.

Un-attested advice is a citation, not accountability. The moment it matters — a plan administrator, an audit, a real clinical fork — someone has to have stood behind it. "The AI said so" is not a defense. A physician's attestation is. AI drafts; a human is accountable.

wall 03

The coaching was never the moat.

The coach is getting cheap because reading your data got cheap — and cheaper still as it moves on-device. So the scarce thing isn't the read. It's the licensed human standing behind the plan: the one dependency you can't self-host.

the layer they can't build

Same data. One physician signature. Everything pre-tax.

You already connected your card, your labs, your Apple Health. AgenticHSA is designed to plug into the exact same sources — and add the one thing an AI agent can't: a licensed physician who reviews each letter and signs or declines it. You approve before a dollar moves. None of this is live yet; here is the design.

1
The agent reads your spend
Read-only, by design. Plaid + Stripe + Apple/Google Pay categorize every health-relevant purchase and flag what may already qualify for pre-tax treatment when a documented condition supports it. It never moves money on its own.
2
It drafts the §213(d) letter
For each flagged purchase, a draft Letter of Medical Necessity in §213(d) format, tied to a real diagnosis on file — or three quick questions if there isn't one.
3
A licensed physician reviews it — this is the part you can't code
A licensed clinician (via altru.care) reviews each draft and signs it — or declines. This is the step no agent, prompt or MCP server can do for you. It's the whole reason this page exists.
4
Dollars route pre-tax — you approve first
A signed letter documents care that already qualifies, so the purchase it covers can be reimbursed from your HSA, pre-tax. Planned: each signed letter gets a hashcare record (timestamp + signature + diagnosis basis) for your audit packet, and your CPA gets a clean year-end ledger.
You built the coach. The signature is the missing piece.
A physician decides. You approve. The 10% of the stack you can't npm install.
where we actually fit
We're not competing with your AI coach. We're building the step after it.

If you were comfortable piping Plaid, your labs and your Apple Health into an agent, you already did the hard part — and you self-selected into exactly who this is for: a builder with cash, no HR department, and a stack of "wellness" purchases that should be pre-tax.

The DIY stack tells you what to fix. A physician's signature is what documents it, when a letter is warranted; what that saves depends on your tax bracket. AgenticHSA isn't live yet. Ask about a spend scan and we'll follow up by email on what may already qualify. Nothing is charged here.

Ask about a spend scan →

Fire the coach. Keep the signature.

Early access is open. AgenticHSA isn't live yet: join the list and we'll email you when it opens. Nothing is charged to join.

Join early access
no w-2 required · nothing is charged to join the list