Reading your words and photos, then explaining calmly.
Petwell is AI-native. The core task is reading an owner's description and any photo, then turning a worry into a calm, plain-language explanation with an honest urgency read and the questions to bring to the vet.
Understanding is the product, not a feature.
Reading words and photos
The way owners describe a problem is messy and human, and a photo carries a lot. Making sense of both needs modern vision and language models, so understanding is the product, not a feature bolted on.
From a worry to a plan
Petwell reads your description and any photo and turns them into a calm explanation, an honest urgency read, what to watch for, and the questions to bring to the vet.
Model-agnostic by design
A thin routing layer picks the right model for the task and can move as models improve. Petwell is not tied to one provider.
How a check flows, end to end.
Four honest levels, with reasons.
Petwell reads every check for how urgent it looks and returns one of four levels. Each level comes with a plain-language reason and what to do next.
This reads as routine. Keep an eye on it and mention it at your next regular visit.
Worth watching over the next day or two. If it gets worse, call your vet.
Worth a call to your vet soon, rather than waiting. Sooner if it is getting worse.
This can be an emergency. Contact an emergency vet now.
The code-level emergency screen runs before the model and can only raise urgency, never lower it. A negation-aware check means “no trouble breathing” does not fire the emergency flag.
Model-agnostic by design.
Petwell is not tied to one AI vendor. It can move as models improve. Your check gets the best available model for the task, not a fixed one.
A thin routing layer
A thin layer picks the right model for the task and can switch without disrupting what you see. As models improve, so does Petwell.
Zod-validated JSON output
Every model response is parsed and validated with a strict schema before it reaches you. Fields must be within length, urgency must be one of four allowed values, and emergency must be a boolean. If the response fails validation, a deterministic template built from the same description stands in.
Code-level emergency screen
A pattern-matching screen runs on every description before the model call, checking for red-flag phrases: trouble breathing, collapse, suspected poisoning, severe bleeding, seizures, a bloated hard belly with retching, a cat unable to urinate, and more. It is negation-aware. A match raises urgency to emergency and cannot be overridden by the model.
No model training on your data
Your description and any photo are sent to a model provider under a data processing agreement. The provider processes them only to fulfil your request. Your data does not train public foundation models.
What Petwell will never do.
These are not design choices. They are rules baked into every layer of the system.
Never diagnoses
Petwell speaks in terms of what it could relate to. It never states a condition as fact. Diagnosis belongs with a veterinarian.
Never gives doses
Petwell will never name a medication, suggest a dose, or recommend a treatment. Those decisions are for your vet.
Always points to a vet
For anything beyond routine, Petwell encourages contacting a vet. For emergencies, it tells you to contact one now.
Care and safety
A full account of what Petwell will and will not do, and how it handles emergency symptoms.
Read the care pageSecurity and privacy
How your data is stored, encrypted, and kept yours. Photos are not retained after a check.
Read the security pageCommon questions about the AI.
See it working on your pet’s situation.
Describe what you noticed in plain words and Petwell reads it, right now. Free to start.