VeriBark is designed around observable canine communication.
VeriBark is an educational tool built on audio and visual behavioral cues that have been studied, documented and taught for decades. The goal is to make those signals easier for everyday dog people to notice and understand.
studies informing 133 communication signals.
The ethogram helps structure visible and audible behaviors so an analysis can connect what appears in a video with relevant canine behavior knowledge.
Behavior is a pattern, not a single cue.
A wagging tail, a yawn or a bark can mean different things in different situations. VeriBark considers combinations of available signals, the surrounding scene, dog profile information and any context or question the user provides.
Multi-agent specialists.
Most AI chatbots look at a video the way a person skims a thumbnail: a handful of low-resolution frames, once. VeriBark processes clips at higher resolution and higher frame rates, because the signals dogs actually use, a lip lick, a weight shift, a half-second of whale eye, a change in bark pitch, are fast, small and easy to miss even for an attentive human.
Rather than asking one model to do everything, VeriBark splits the job across specialist AI agents, each group with its own responsibility:
Observation
Agents that watch and listen frame by frame: face, ears, eyes, posture, tail and movement on the visual side; barks, whines, growls, pitch and timing on the audio side. Its job is to record what is there, not to decide what it means.
Cross-reference & interpretation
Additional agents match those observations against the 133-signal ethogram and its evidence base, weighs them together with the context you provided, and explains the reasoning behind the interpretation.
Separating observation from interpretation keeps each step focused and checkable, so a subtle cue caught in the first pass is not lost in the second.
How accurate is it today, and where is it heading?
In our current internal evaluation on labeled training videos, VeriBark's interpretations align with the expected assessment approximately 83% of the time. That is an early indicator, not a guarantee for every video: dog behavior is contextual, and lighting, camera angle, sound and the information you provide all affect an analysis.
The model is retrained continuously on feedback from users and from canine behavior experts, and the ethogram and test suite grow with it. Accuracy is a moving number, and the direction is up.
Examples from the evidence base.
VeriBark compares what it sees and hears in a video against established frameworks developed by veterinarians, animal behaviorists and researchers, including gold-standard canine ethograms and standardized behavioral assessment models. These are several of the major frameworks, research programs and references represented in that work.