A Palo Alto startup is betting that the most useful health sensor in your home is the thing your dog already visits twice a day

In this issue: Hoomanely comes out of an 18-month beta with 5 million data points, a $29/month feeding station, and a plan to become the health data layer for animals. Here’s what’s real, what’s unproven, and what it signals about where ambient AI sensing is headed.

The setup

Dogs are terrible patients. Evolution taught them to hide weakness, so by the time a limp or a slump in energy is obvious enough for an owner to act on, the underlying problem is often weeks old and considerably more expensive to treat.

What dogs cannot easily fake is how they eat and drink.

That is the entire thesis behind Hoomanely, a Palo Alto company that this week unveiled an AI-native platform for animal health built around a deceptively boring piece of hardware: a feeding station.

What they built

The product is called EverBowl. Underneath the food and water bowls sits a sensor stack that passively logs consumption volume, eating speed, chewing and swallowing audio, facial thermal patterns, and oral motion during each feeding event.

Edge machine learning fuses those signals in real time and compares them against one reference point only: the same dog’s own history. There is no population average being applied here. The system learns what normal looks like for your animal, then flags sustained departures from it through a companion app and longer-horizon reports designed to be handed to a veterinarian.

The owner can also feed context back into the model — things the sensors cannot see — so the baseline stays honest.

The design logic is worth pausing on. As CEO Sai Supriya Sharath framed it to TechCrunch, the bowl is close to an ideal capture environment: the same location, the same posture, the same routine, twice a day, for years on end. Most consumer health sensing fails because compliance decays. Nobody has to remember to use a food bowl.

The evidence so far

An 18-month beta produced more than 5 million multimodal data points across 80-plus dogs.

Two cases the company highlights:

• A shift in one dog’s eating pattern traced back to a chipped tooth that was beginning to become infected — the kind of dental problem that runs into four figures once it advances.
• A sustained deviation from both eating and temperature baselines that pushed an owner to book a vet visit, where the dog was diagnosed with tick fever.

The platform also tracked signal changes across a course of treatment for a dog with Cushing’s syndrome.

Now the honest part

This is a small cohort, roughly 50 devices shipped, and a company that will tell you plainly what it does not yet know.

Hoomanely has no independent sensitivity, specificity, or false-positive figures for clinical events. Producing those requires formal study design — comparing what the system flags against what a veterinarian actually diagnoses — and that work is being scoped now. Sharath has said results will be reported per alert category rather than compressed into a single marketing-friendly accuracy number, which is a more defensible position than most consumer health hardware takes.

The company is also explicit that this is not a diagnostic replacement. It is a trigger for earlier action and a richer data packet for the vet who eventually sees the animal.

Pricing is $29 per month for the device plus app reporting.

The actual business

Hardware is the acquisition channel, not the destination.

The roadmap adds EverSense, a wearable for movement and rest data, and EverHub, an aggregation layer that ingests third-party inputs — smart collars, feeders, home environment sensors. Each addition thickens the longitudinal record on a given animal.

The long game Sharath describes is becoming an animal health data company: a proprietary, continuously refreshed dataset that supports veterinary research and serves nutrition and insurance buyers. And because the capture architecture is not dog-specific, the same approach could extend to cats, horses, or livestock — where the economics of early detection at herd scale are considerably larger than anything the consumer pet market offers.

Funding to date: $1.8 million pre-seed, with seed conversations underway.

Why this matters beyond pets

Three things here generalize well past the dog bowl.

1. The best sensor is the one nobody has to adopt. Wearables and tracking apps lose to attrition. Hoomanely instrumented an existing, unbreakable daily ritual instead of asking anyone to change behavior. If you are designing an AI product, the question worth asking is which routine in your user’s life is already perfectly repeatable — and whether you can put sensing there rather than adding a step.

2. Personal baselines beat population models. The value is not in knowing what an average Labrador does. It is in knowing what this Labrador does, and detecting drift. That is a defensible data position that compounds with time and is genuinely hard for a fast follower to replicate.

3. On-device processing as a positioning asset. Video is analyzed at the edge and discarded rather than stored or transmitted. For a product that involves a camera pointed at a domestic space, that is not a footnote — it is the reason the sale closes.

Everything rests on the clinical studies. A system that surfaces real problems weeks early is transformative for both owners and veterinarians. A system that generates a steady drip of false alarms trains people to ignore it, and quietly becomes worse than no system at all.

The company knows this, which is a reasonable sign. Watch for the per-category results.