Verifiable end-to-end encrypted inference with @PayPerQ just landed in getbased.

So yes, you can now try GLM 5.2 privately.

No email.
No KYC.
No fiat.
No subscription.

Just pay per query and keep your prompts encrypted.

Biological Coherence: the missing layer in lab interpretation


What follows is not poetry, but a reality for many people trying to debug and solve their health issues.

Some markers are green.
Some are red.
The system is still failing.
And you have no idea why.

This quote from The Expanse nails it hard.


Artificial ecosystem, you ask? That's right.

We do not realize it, but our environment resembles almost nothing of what our biology evolved and adapted to.

So the less we comply with Nature, the more we need to use technology to fix what was not meant to be broken in the first place.

Which is ironic, because our technologies are what makes us sick.

And if you look at a map of biological pathways, you quickly realize that whatever you are about to do, there is a pretty good chance your plan has holes and leaks all over the place.

Biology scores, the human-readable puzzle

Markers from blood, urine, and other tests can be useful, but only if:

a) you can interpret them right
b) the interpretation is context-aware

Neither is easy.

Hell, for most people it is almost impossible.

This is where state-of-the-art AI can help. If it is good at one thing, it is making sense of a big chunk of incoherent data thrown at it.

getbased just got Biology Scores, a first iteration of a new scoring system. A user-facing, readable format that computes and interprets your health data into something that can tell you what is working well and what is broken.


Biological Coherence is an overview score composed of up to 13 baseline scores and 5 extended scores.

Each score has two levels of inputs: core markers, as the bare minimum, and extended markers for better confidence.


Take Metabolic Flexibility, for instance.

The viable minimum is:
fasting glucose
fasting insulin
triglycerides
HDL

For better confidence, add:
HbA1c
C-peptide
fructosamine



Every subscore will not just show you a meaningless number. It will explain what it means, what drives it, what drags it down, and most importantly, how it is tied to your other data.

That includes DNA/SNPs and mtDNA, lifestyle data like diet, sleep, and stress, environmental context like light and nnEMF, medical and family history, and biometrics like HRV, heart rate, blood pressure, weight, BMI, and more.

Scores are context-aware.

Meaning if, for whatever reason, creatinine levels are not relevant for you, getbased will not count it and penalize you for low eGFR while screaming that your kidneys suck.

To make the whole thing easier, there is a Coverage Planner that helps you pick what to test next so the scores become more complete and actionable.



We all know that compiling the test list without leaving anything important out, and without wasting money on nonsense, is the real pain in the ass.

getbased just got a bit more powerful, and with user-facing scores, hopefully more useful for everyone, including non-geeks and non-nerds.

The best part of getbased is that you can just talk to it, and it will explain everything.

Go try it with female or male demo data: https://app.getbased.health

As always, feedback, issues and PRs are welcome: https://github.com/elkimek/get-based

And in case you want private no-KYC inference, both @routstr and @PayPerQ are supproted

Bunch of import-related improvements landed recently while working on a new major feature.

New Lens is coming, stay tuned!

Imports now support .csv and .xlsx

Still kicking, clearing tech. debt and improving the codebase.

Bunch of awesome features getting ready. Stay tuned.

There's been a lot of ongoing work behind the scenes over the last three weeks - optimizing the code, fixing bugs, and polishing things up.

But here come two new super useful features:

1) Reports - Generate a printable PDF for your practitioner or just for yourself.

2) Share your profile via a link and password. Everything is encrypted, so only the person who receives it can see your data.

Refactoring 5 days in a row. Back to creative work soon, hopefully.

getbased v1.6.0 — Photobiology release

The dashboard now opens with a Light Today card: a one-glance read on your day's exposure.

UVI, air quality, six biological light channels, and one-tap entry to start a Sun or PBM (photobiomodulation) session.



Open Light & Sun drops you into the dedicated photobiology view.

Conditions first: UV Index, sun position, ozone column, air quality.

Below it, today's sun arc — sunrise, first UV-A, midday UV-B peak, UV-A fall-off, sunset.

All data comes from a self-hosted Copernicus Atmosphere Monitoring Service (CAMS) relay you can run yourself: https://github.com/elkimek/getbased-uvdata



Light Setup asks a handful of questions that anchor the math: skin type, home lighting, eyewear outside, baseline light burden.

It's what lets the model compute your vitamin D synthesis, eye + skin saturation, minimum erythemal dose, burn rate, and the rest of the sun-exposure picture honestly — not generically.

From here you can also start a sun or PBM session.



Your light, by what it does — six channels, computed from every sun and PBM session:

- Vitamin D
- Body clock
- Cellular repair
- Cardiovascular
- Mood & hormones
- Outdoor eye light

Behind them sit the mechanisms light actually drives in your biology: circadian entrainment, melatonin synthesis, nitric oxide release, the POMC system, and dopamine — pathways that are exclusively (or near-exclusively) controlled by light.



Tap any channel for a detail view: how saturated it's been over the last 7 days, which days banked it, what would tip it up next.

Beneath the chart, an AI overview ties it back to what you actually did — and what to do about it.



Before a sun or PBM session you tell it which body parts are uncovered. That drives correct channel math (e.g. how much vitamin D you actually synthesized — face-only is not torso).

It also flows into AI context, so later it knows whether you're skipping the parts that matter — thyroid, breast, genitals.

Basically: show us on the doll where the sun touched you.



When a session ends you get a per-channel breakdown — how well you lit up each one, plus an analysis.

Vitamin D math is intentionally conservative and weights your DNA (more on that soon), so you won't see "10 minutes of midday sun = 10,000 IU" fantasy numbers.

The maths behind it: https://app.getbased.health/docs/contributor/sun-spectrum-model

That covers sunlight.



PBM devices: pick from curated brand presets (Mitochondriak, Chroma, EMR-Tek) or drop a product link and the AI fetches the spec sheet for you.

From there a PBM session works the same as a sun session — body parts, timer, channel readout.



Indoor light shapes you more than most people realize, and unless you measure, you genuinely don't know what you live under.

Rooms you spend time in lets you build out each room — light source, hours per day, displays (including phones and tablets you carry between rooms). Eight on-device tools help you measure flicker, CCT, brightness, and spectrum from your phone.

The AI then reads everything together and tells you how bad it is — with concrete fixes.



Snapshot a room as an audit. Change something — swap the bulb, fix display settings, kill a PWM driver — and snapshot it again.

Side-by-side, before/after, with severity badges per room (lux, flicker, CCT, melanopic).



Light Tools. Your phone's sensor has real limits, but internal testing says it's good enough for "test, don't guess."

- **Quick checks (10–30s):** What is this light? · Lux Meter · Color Temp
- **Full measurements (30s–2min):** Flicker Detector · Sleep Darkness · Window check
- **Walkthroughs:** Home audit · Golden hour log

Most people have zero idea what light they live under or what it's doing to them. Now you can find out in an afternoon.



24K lines of code, 25 modules, 250 commits, 9 days of work — first version is out.

Expect bugs, rough edges, and the occasional weird UI moment.

If light is your cup of tea, take it for a spin and file what breaks on GitHub.

getbased started as a "nice charts for my blood work" and evolved into Personal Health Intelligence.

When you let your personal agent tap into your data, you'll see a magic happen. Health is just an engineering problem, so let's solve it piece by piece.

Biometrics are here.

How does your resting heart rate, HRV, blood pressure, and the rest of your wearable data related to your lab work, lifestyle, DNA, and a bunch of other aspects nobody has yet linked together?

Fuck around and find out.

👉 https://getbased.health

Bugfixes & Improvements ™

getbased just got orders of magnitude more powerful.

Imagine you live right next to the Library of Alexandria.

Every piece of knowledge in the world is right there, waiting.

But how do you even begin to find what you truly need?

Now imagine your best friend works there as the librarian.

She's not ordinary. She was born with this incredible mind. Her brain was built differently.

She sees every book, every page, every important paragraph and gives it its own unique number.

She's quietly numbered everything that matters.

When you ask her a question, she doesn't search.

She just hears the numbers.

And in a fraction of a second, she pulls exactly the passages you need — the ones that actually fit.

That's RAG. And it works because meaning has geometry.

Every idea lives at a coordinate in a space no human eye can see — but she can. She maps your question to the same space and finds what's nearest. The laws of math do the rest.

Here's how it feels in getbased:

You build your own knowledge base — your research papers, clinical guides, the health frameworks that changed how you think.

Whatever truly matters to you.

When you ask getbased something about your labs or your health, the model first turns to her:

"What's relevant?"

She hands back the exact pieces that belong.

Only then does it answer — grounded, personal, real.

Your librarian.
Your library.
Your health story.

➡️ http://getbased.health

Opus 4.7 (Sonnet 4.7 soon) or Rosalind?

Any guesses which will be better?