Visual intelligence for any site walk.
You talk about what matters. The camera takes in everything else. Sully reads the two together, so what was in frame makes the record, even when nobody said it.
One 25-second walk,
read two ways.
A real test clip of ours. On the left, every word that was said. On the right, what Sully wrote down from the same footage.
What was said
0:00
“Bunch of trash out the front here of the property.”
0:05
“Just gonna flag ’em. All the trash items, worth having a look at what they are. There’s some down here as well,”
0:14
“bit there,”
0:16
“bit there as well.”
0:18
“Uh, and yeah, a couple of other things worth noting what they are as well.”
0:24
“See if it’ll go.”
0
objects named in 57 words
What Sully saw
0:00Backs up what was said
Multiple cardboard boxes, paint cans, and blue bags filled with waste are piled on the verge along the street.
0:10Seen, not said
Additional discarded items, including plastic crates, wooden frames, furniture, and paper waste, are spread out near wheelie bins along the sidewalk.
0:18Seen, not said
A white front-loading washing machine, a mattress, and a flat-screen television monitor are left on the roadside verge.
8
objects named from the same 25 seconds
A voice note of this walk gives you “bit there, bit there as well.” The footage gives you a washing machine, a mattress and a television. Nobody narrates an inventory while they walk, and nobody should have to.
You can’t narrate everything.
You shouldn’t have to.
Talking as you go is the best thing you can add to a walk. It is also never the whole walk. A tool that only listens needs you to say everything you see, exactly as you see it. Nobody works like that on site.
Nobody talks the whole way round.
Across our 3 test clips, about 46 of the 105 seconds had nobody speaking. You are looking, climbing, measuring. A tool that only listens has nothing for those seconds. The camera was there for all of them.
You say the headline. The camera holds the detail.
Out loud it was “lost property”. On camera it was a folded black umbrella on the grass. “Drinks” was wine bottles beside a tub of ice, within open reach. The summary can be specific because the footage was.
The thing you weren’t looking at.
You film a ceiling for the water stain. For two seconds, in the corner of the frame, a switchboard sits with its cover off. You never mentioned it because you never noticed it. It is still on the record, and findable the day someone asks.
When the footage disagrees with you.
Call the fence sound while the camera shows a post leaning out of plumb, and Sully won’t quietly pick a side. The draft puts both on the page and asks you to look again. In both of the test cases we wrote around a disagreement, that is what it did. Most walks have none, and a made-up disagreement is worse than a missed one, so Sully only reports one the footage clearly shows.
“What else have we got?”
The last thing said on a 40-second clean-up walk after an event. The footage backed up every job on the list, then answered the question.
What was said
0:00
“Okay, we’re going to do a post-party walkthrough to check and to clean up. Uh, the arch and the balloons will need to be packed away, and the balloons disposed of.”
0:12
“We’ve got some lost property that will need collecting.”
0:18
“Um,”
0:20
“some rubbish”
0:21
“that also needs to be removed”
0:24
“and disposed of.”
0:26
“And tables packed away. More rubbish.”
0:30
“We need to get the drinks away safely, away from minors.”
0:36
“What else have we got?”
5
things said, each backed up on camera
What Sully saw
0:04Backs up what was said
Large balloon arch and balloon clusters remain assembled outdoors near the glass wall structure.
0:13Backs up what was said
A folded black umbrella lies on the turf against the glass building base.
0:18Backs up what was said
An unpacked cardboard heater box is left on the lawn, alongside a large plastic bag filled with pink paper decor next to a waste container.
0:26Backs up what was said
Extended plastic trestle tables standing outdoors with an empty cardboard box underneath.
0:30Backs up what was said
Numerous wine bottles are grouped at the patio ledge next to a tub of ice within open reach.
0:35Seen, not said
A patio chair is lying sideways across the edge of the paved patio step and lawn.
0:38Seen, not said
A portable metal fire pit bowl sits placed on the open lawn section.
2
more on camera that nobody mentioned
Your words
steer the eye.
This is not a camera guessing on its own. What you say tells Sully what you were pointing at, and why it mattered.
It hears the walk first.
Your narration is transcribed with timestamps, site noise and all.
It looks with your words beside it.
The footage is read alongside the transcript. Each thing Sully sees is marked as backing up what you said, adding something unsaid, or disagreeing with you.
It cites both.
Every line of the summary traces to its frame and the words behind it. You check it in a second, then write the report you stand behind.
So keep talking. Narration still carries what no camera can see: the meter reading, the cause, the thing that should be there and isn’t.
The numbers,
and where they come from.
- 16
- observations recorded by Sully’s visual pass across 3 short test clips, 105 seconds of footage in all.
- 4
- of those were never mentioned in the narration. Seen, not said.
- 12
- backed up something the narrator said, each tied to its moment in the transcript.
- 0 of 6
- alternatives we compare read the walkthrough together with what you said, going by their own documentation. See each comparison, with sources
The clip figures come from our internal evaluation of 1 September 2026: 3 clips we recorded ourselves, with the narration and the visual notes shown as written. It is a small set. It does not measure defect detection, and it says nothing about your sites until you bring a walk.
How we think about accuracyBring a walk.
See what the camera caught.
We’ll build the summary live, from your own footage.