Training with Coach

The button under Start Training on any plan page. It starts a conversation with Coach about the session you are about to do, and it does not commit you to anything.

It exists for a specific person: the one who would like a trainer in the loop and has nothing in particular to report. You do not need a problem. The button sends Coach the plan and the step you are on, so the first thing you get back is about that step rather than "how can I help".

What Coach does next is post a card. Nothing opens on its own. The card either opens the plan in the ordinary Player, with Coach checking in as you go, or it opens the camera for a live session where a model marks the reps. Coach picks which, out loud, and tells you which one it is picking and why before the card appears. If it guessed wrong, say so and ask for the other one.

The Player with check-ins is the default, and it is the right answer for most plans, all quiet work, anything to do with fear, and any step you are still teaching. See Which Mode a Session Should Get.

Often confused with

Start Training. The green button above it opens the plan on its own, which is the ordinary way to train and always will be. This one brings someone along.

The ordinary Player, with Coach looking up now and then. You judge every rep yourself, exactly as you always do.

Coach does not watch every rep. A remark on each one would be worth nothing and would spend your energy on saying so. Instead it arms a small set of conditions when the session starts, and the phone wakes it when one of them fires: after so many reps, after so long, when the session reaches a particular step, or when the engine does something interesting. A drop or a split is usually on the list, because those are the moments worth a sentence. When Coach is woken it gets the reps since last time, and it can say one short line out loud. Then it arms the next one.

You can also just talk to it. Hold the mic in the Player and say what is happening. "She is taking the treat but not sitting first." "Was that a pass?" You get a spoken answer back, one or two sentences, without putting the dog down or typing with a handful of chicken. This is the part testers keep describing as the fun part, and it is the part that would not work as text.

Two more things ride along with a check-in when you turn them on:

The room, transcribed. With Listen on, the audio since the last check-in is transcribed and goes to Coach with the reps. That is the only route by which Coach can ever know that a cue got repeated three times, that your marker is landing a beat late, or that the word drifted from "off" to "leave it" halfway through. Numbers cannot show any of that.

A clip. You can record a short video of a rep from inside the Player and it reaches Coach with the next check-in. It is being reviewed, not scored: hand height, lure path, where the reward lands, what your dog's body is doing. See Recording a Clip for Coach.

Coach staying quiet is not a verdict. It looks when a condition it set fires, so silence means nothing has fired, not that everything is fine. You are still the one watching your dog.

Often confused with

Scheduled Check-ins in the other sense, the scheduled kind that run while you are away from the app entirely. Same word, different feature.

The other mode, and the one still in beta. A model watches and listens through your phone, marks each rep Pass or Fail, and talks to you while you work. You still set the dog up, prompt the behavior, and pay. What you hand over is the tapping, and what you get for it is both hands free and no reaching for a phone at the moment your dog is doing the thing.

That is the whole case for it, and it is a real one: reaching for a phone costs mark timing. The rest of this entry is what beta means in practice.

Coach decides whether a session gets this mode and says so first. When it does, the card you tap opens the camera and waits until it can see both you and your dog before anything starts, so a session never begins pointed at an empty room. That waiting screen is also your consent moment, and it is meant to be unmissable: the frames and the audio leave your phone and go to a server, because the model is far too large to run on one. A live session is the one part of the app that cannot work offline.

What happens to the video and the audio. Both stream to a GPU server we operate. The audio is transcribed there and is not kept as audio; what is saved is the text of what you said, the frames the model saw, what it said back, and a manifest of the session. That log is kept for as long as your account exists. One thing there that surprises people: the microphone keeps streaming while the live trainer is talking, so the room is transcribed continuously rather than in turns.

The log is also one of the things we use to train the live trainer's model, and the Privacy Policy states that "there is no in-app toggle to exclude your data" from that. You can be left out of future training runs by emailing [email protected]. Whatever you point the camera at gets recorded, other people in the room included, and the consent for that is yours to get. The whole of it is in What Is Kept, and What Trains the Models.

Prop the phone somewhere that keeps the whole working area in frame, including the spot your dog leaves to and comes back from.

Coach does not go away while this runs. It gets woken at points it chose in advance, and can say one short line out loud if something about your dog's welfare or your criteria needs saying. Everything longer waits for the chat afterwards, where the session gets debriefed properly.

What does not change is who decides difficulty. Push, Drop, Stick and Split still run on the verdicts, exactly as they do when you tap them yourself. The live trainer is a pair of eyes. It is not a coach, and it is not a judge of your dog.

Often confused with

An automatic trainer. It marks reps. You train the dog. If you were hoping to put the phone down and leave the room, this is the wrong feature and there is no right one.

This applies when the live trainer is marking. It is the one thing the beta asks of you, and it is worth more than anything else on this page: use a marker word, and either use a no-reward marker or tap the X yourself.

The reason is mechanical. Your voice is transcribed, and the marker word is what makes the model stop and look. It can call a rep without one, but less reliably. With one, it is genuinely good.

So, in practice:

Mark the pass. A click or a word, the moment the behavior happens. See Marker. Plenty of good training does without one, and position feeding in particular needs no marker at all. This mode does, because the marker is how the phone knows a rep happened.

Then either a no-reward marker, or the X. A no-reward marker is a flat, cheerful "Oops" or "too bad" that means no treat this time, and it gives the model the clearest possible signal on a miss. If you would rather not use one, or you are working with a dog for whom it is the wrong idea, tap the X on the phone yourself instead. A rep you mark by hand is scored exactly like a rep the model marks. Mixing the two mid-session is fine.

Silence is the case that goes wrong. A quiet rep with nothing said and nothing tapped may simply not be scored.

None of this is asked of you in the Player. There you are the judge, and a marker word is only ever for your dog's benefit.

Often confused with

Talking to the model. You are not giving it instructions, you are training your dog out loud. Cue, mark, pay. It listens in.

About two frames a second, through one fixed camera, with the audio alongside.

That resolution is honest about some things and blind to others. Sustained, whole-body information reads reliably: where the dog is, whether they are standing or lying down, stiff or loose, oriented toward you or away, in frame or gone. Brief small signals do not. A lip lick, a blink, a yawn, a paw lift lasts a fraction of a second and will often fall between two frames entirely.

So it does not tell you how your dog is feeling, and it is not built to. Nothing at two frames a second earns a sentence about how a dog feels. It marks reps.

Which means its quiet is not a green light. The camera not objecting is not the same as your dog being fine, and nobody is watching your dog's body language for you. If a session feels wrong, end it; you never need the phone's agreement.

It can also stop on its own. A live session depends on a network and a rented GPU, so the voice can arrive late, get cut off, or drop out entirely, and the session can end without warning mid-set. Nothing is lost when that happens: the reps already scored are in your session like any others, and you can carry on in the Player by hand or tap the card again. The rule it points at is the one worth keeping anyway, and the Terms of Service put it well: stop the exercise yourself whenever the guidance does not match what you are seeing. A dog set up and waiting while you troubleshoot a phone is a dog learning that the exercise means nothing.

Often confused with

A welfare monitor. It is a rep marker. Nobody should be leaving a dog in a situation on the strength of a camera not objecting.

Nobody gets refused a session. But plenty of sessions should not be handed to a camera, and Coach is the one routing them, so it is worth knowing what it is routing on.

The live trainer needs a behavior that is marked, active and visible: a retrieve, a stay, a position, a recall, something that happens at a moment you can hear. Take any of those three away and it has nothing to work from.

Which rules out, for now:

Counter-conditioning and other silent work. A Classical step, and a lot of Errorless work, is deliberately wordless: you are pairing something the dog finds uncomfortable with food, and there is no marking moment to hear, because there is no behavior being marked. We have tested it, and it misses most of the reps: not because it judges badly, but because from outside, a dog being fed in silence and nothing happening at all look the same.

Fear and anxiety plans generally. Same reason, plus a harder one: these are the sessions where the decision that matters is when to stop, and that decision is not one a camera gets to make.

The first time you teach something. Early reps are where criteria are messiest and where you learn what your own pass actually looks like. Handing that over before you have decided it yourself is a way to end up with a plan full of verdicts you do not agree with.

All three go to the Player with check-ins instead, which is not a consolation prize. Coach in your ear while you judge your own reps is the better tool for every one of those cases, and for a good many of the others.

And it is the ordinary thing, not the awkward thing, for a trainer to want to watch their own dog rather than a phone.

Often confused with

A judgement about your dog. These are limits of a camera and a microphone, not of the dog in front of it.