An AI observer in every stall — one that learns each horse's individual normal and alerts you the moment behavior drifts toward colic, stress, or injury. No video leaves the barn. No one has to watch a screen.
In a typical stable, horses are checked every 6–12 hours overnight. A horse that begins showing early signs at midnight may not be examined until morning — by which time a manageable medical case has become a surgical emergency.
Before a condition becomes clinically visible, it leaves a trace in behavior: subtle shifts in movement, posture, and activity rhythm. Those signals are continuous, subtle, and highly individual — below the threshold of reliable human detection. The information exists. Until now, nothing was built to read it.
Each Observer Agent runs its intelligence on the device itself. It watches one stall, learns one horse, and reports in structured events — not video streams. Four layers turn observation into action.
AI inference on-device, around the clock. Raw footage is processed and discarded in the barn; only structured events and short context clips move on.
Signals from every stall combine into individual baselines per horse — and context: unusually still, or simply asleep? The system learns the difference.
Hours- and days-long patterns feed early-risk scoring for colic and stress — flags designed to surface before clinical symptoms appear.
Not just an alarm: what was observed, over what period, why it's abnormal for this horse, and what to do — delivered to owner, trainer, and vet.
| What it watches | How | Why it matters |
|---|---|---|
| Rolling behavior | Continuous pose tracking against baseline | Primary behavioral indicator of colic |
| Extended inactivity | Activity deviation from the individual norm | Early sign of systemic illness or pain |
| Abnormal movement | Gait and locomotion pattern analysis | Colic, lameness, neurological events |
| Posture anomalies | Position relative to this horse's baseline | Pain, respiratory distress |
| Feeding deviation | Absence from feed at expected times | Systemic, dental, or gastrointestinal issues |
| Respiratory patterns | Flank-movement frequency analysis | Respiratory infection, pain response |
Every bounding box, skeleton point, and readout below is generated on-device, in real time — individual pose confidence, orientation, lying probability, and estimated standing duration, per horse, per frame. This is the gap between a motion-triggered camera and an agent that actually understands what it's looking at.
The Observer Agent unit isn't a relabeled consumer security camera. It's built around an on-device AI chip powerful enough to run full pose estimation, per stall, continuously — with nothing streamed to the cloud to make it work.
| Component | Spec |
|---|---|
| Image sensor | 4K UHD (3840×2160), 1/1.8" CMOS, 2.0 µm pixels |
| Field of view | Wide (109.8°→46.2°) or telephoto (44.5°→14.5°) lens option, ~4× optical zoom |
| Low-light / IR | <0.01 lux full-color night vision, auto IR-cut, HDR, gyro-stabilized |
| On-device AI chip | Dedicated edge AI accelerator — up to 20 TOPS (INT8), <50 ms inference, multiple models running concurrently on-device |
| Local storage | 64 GB onboard + expandable microSD |
| Connectivity | PoE, single-cable install; structured events over MQTT/REST — no continuous video upload required |
| Housing | IP67 (dust/water) + IK10 (impact), fanless, −40°C to +60°C operating range |
| Power | PoE (802.3at) or 12V DC, ~5–6W typical draw |
This is enterprise-grade edge AI hardware — not a repackaged consumer security camera. Full spec sheet available on request.
Barn cameras aren't new. What's in them is.
Streams raw video to a server, all day, every day. Someone still has to watch. A server failure blinds the whole barn.
Detects predefined events against one-size-fits-all thresholds, then uploads clips to the cloud. No individual baselines. No coordination.
Fully on-device AI with individual baselines, peer coordination between agents, and structured intelligence instead of raw footage.
The system never compares your horse to a breed average. It learns what normal looks like for that specific animal — and alerts on deviation from her own baseline. Fewer false alarms; earlier true ones.
Wearables count steps and heartbeats. Observer Agent reads behavior — the same signals an experienced horseman reads — continuously, and without fatigue.
No video stream ever leaves the barn. Footage is analyzed and discarded on the device; only structured events and short context clips are transmitted.
Every agent operates independently. A Wi-Fi outage at 2 a.m. does not create a detection gap — the stall is still being watched, and events queue until the connection returns.
Each plan pairs a purpose-built Observer Agent camera unit with a monitoring subscription. Because facility size, horse count, and hardware scope vary widely, we scope pricing per engagement rather than publish a flat rate — tell us about your barn and we'll come back with numbers.
A single colic surgery — or the loss of a horse — costs many times what a season of monitoring runs. We'll size a plan, or a partnership, around what your barn actually needs.
For high-volume stables, racetracks, and breeding operations, we go beyond per-horse pricing — structuring joint-venture and revenue-share partnerships with a guaranteed monthly minimum per facility. Built to scale with your operation, not penalize you for its size.
Discuss a partnership24/7 watch over every horse, early intervention on colic and injury, and objective insight into recovery and rest.
One platform from a single stall to an entire track. Reduced incident liability and a visible welfare credential.
Arrive with a 72-hour behavioral log, a deviation score, and the triggering clip — instead of reconstructing events from memory.
Verified monitoring reduces mortality and surgical claims; structured behavioral logs support claim validation and pricing precision.
We also structure certification, co-branded distribution, and strategic programs with breeder & owner associations. Talk to us about a partnership →
Edge agents live in the pilot barn: detection of abnormal movement, rolling, and inactivity, with the first structured behavioral dataset.
Veterinary feedback loops, individual baselines per horse, and configurable anomaly scoring — accuracy that improves measurably with use.
Temporal models designed to flag risk before visible symptoms: early colic probability, stress trajectories, vet & insurance integration.
Cross-facility benchmarking and population-level insight — a behavioral intelligence layer for the equine industry.
We're enrolling a limited group of U.S. pilot facilities — professional stables, breeding operations, and training centers — plus partners across veterinary, insurance, and racing. Tell us about your barn, and we'll come back with a plan.
We're putting Observer on our own family's farm first — before we ask any client to.
Open the pilot applicationInvestor & partnership inquiries welcome. Horse Observer Agent is a venture of Newton Ventures — a physicist-led AI engineering firm, est. 2010.
