After three sessions with OttnoAI — one long session, one that disappeared when I closed the tab, and a third that produced a long set of recommendations — I’ve come away with a clear sense of what this tool is, what it isn’t, and why I’m not going to subscribe at $20/month.
This isn’t a negative review. I’m genuinely glad I tested it. I think the founder is building something interesting, and I appreciate the privacy‑first stance. But the product, as it exists today, isn’t something I can justify paying for — especially when the core value should be built directly into Garmin Connect or even MyFitnessPal.
Here’s why:
1. Every session starts from zero — no memory, no continuity, no way to save
OttnoAI has no way to:
save a chat
resume a session
pick up where you left off
maintain context across conversations
If you close the tab, the entire session is gone. My second session — which included a long, detailed back‑and‑forth — simply vanished.
This means you have to:
copy/paste everything into Word or Notes
re‑explain your context every time
get very good at prompt engineering
manually reconstruct your own history
For a tool that’s supposed to help interpret long‑term health data, starting from zero every time is a major limitation.
This alone makes it hard to justify a subscription. I get the same level of service from the post-Amazon acquisition One Medical level of care by physician assistants acting as primary care practitioners.
2. The trial banner never updated — the UI feels unfinished
For three days straight, the banner at the top of the screen said:
“3 days left in your trial.”
It never updated unless I manually refreshed the window.
It’s a small thing, but it signals that the UI is still early and not fully wired up. Combined with the typing lag and occasional freezing, it reinforces the sense that the product is still in a prototype phase.
3. The hallucinations are frequent, and sometimes stubborn
I expect hallucinations from any LLM — that’s not the issue. The issue is the type of hallucinations and the fact that some persisted even after correction.
Examples:
It told me to “contain the cats” (as if that’s ever happening).
It invented a “coursework intensity timeline” out of thin air.
It repeatedly insisted my L4‑5 spinal fusion was causing ongoing pain — even after I corrected it multiple times and explained that the surgery solved the problem completely.
It assumed my midterm project was an “exam day” and blocked it out as a rest day.
It confidently told me it didn’t have my 30‑day data… until I uploaded the CSV… at which point it said, “Oh yes, I do.”
These aren’t edge cases. They happened in every session.
The model did correct itself when prompted, but the fact that it needed repeated correction — especially about the spinal fusion — is a sign that the grounding and guardrails aren’t strong enough yet.
4. It overreaches into medical interpretation
This is where I get cautious.
OttnoAI drifted into:
diagnosing causes of heel pain
predicting recovery timelines
interpreting autoimmune interactions
prescribing caloric deficits
making claims about hormonal patterns
giving sleep‑architecture interpretations that sounded authoritative but weren’t grounded in my actual data
I understand the founder’s intent that this is meant to be a supportive tool, not a medical device. But the model’s tone sometimes crosses that line, and users may not always know when it’s guessing.
This is exactly why I think this kind of tool needs stronger constraints before it’s ready for a paid tier.
5. The helpful recommendations were good — but not $20/month good
To be fair, I did get a few genuinely useful insights:
Stop taking melatonin every night as it’s more disruptive than helpful
Add brown noise to my nightly routine
Take progesterone at the same time every night (8:30–9pm) and give it 60–90 minutes to work
Try a 5‑minute box breathing exercise after driving
Warm the bed with the heated mattress pad, then turn it off when I get in
These are small, actionable, grounded suggestions which are exactly the kind of thing Garmin should be surfacing.
But here’s the thing:
These insights came after hours of prompting, correcting, and steering the model back on track. They weren’t the default output. They were the result of me doing the heavy lifting.
That’s not a $20/month experience.
6. This functionality belongs inside Garmin Connect (or MyFitnessPal), not as a standalone subscription
Garmin already has:
the data
the sensors
the long‑term history
the stress and HRV models
the sleep architecture
the recovery algorithms
What they don’t have is the interpretation layer — the connective tissue that helps people understand why their sleep tanked, why their stress spiked, or why their heel hurts after certain activities.
OttnoAI is trying to build that layer. But it shouldn’t require:
exporting CSVs
manually uploading files
re‑explaining your context every session
paying $20/month for something Garmin could integrate natively
This is the kind of functionality that should be built into Garmin Connect or MyFitnessPal as part of the existing ecosystem, not a separate subscription.
7. I’m glad I tested it — but it’s not ready for me to pay for
OttnoAI is ambitious. It’s privacy‑first. It’s trying to solve a real problem. And I genuinely appreciate the founder’s approach.
But the product today:
loses sessions
hallucinates frequently
overreaches medically
misinterprets context
lacks grounding
has no memory
requires constant correction
feels like a prototype
and doesn’t yet deliver $20/month of value
I’m rooting for it. I want it to succeed. But right now, it’s not something I can justify paying for — especially when the core value should be built directly into the platforms that already hold my data.
No comments:
Post a Comment