Detect and track, search your whole library in plain language, ask an agent across every clip, extract structured data, and generate reports — the intelligence was always in the video. We read it out.
Semantic search reads what's happening across every clip you've indexed — then an agent answers questions and cites the exact timecoded frames, wherever they live.
forklift within 2m of a person, no hi-vis vest
warehouse_B · aisle 4
forklift + person · 1.4m gap
00:02:41
dock_cam_03
no hi-vis · confidence 0.91
00:11:07
warehouse_B · aisle 1
forklift + person · 1.9m gap
00:18:52
Did any vehicle block the fire lane yesterday, and for how long?
Ask a question and land on the answer. Read a hundred clips in one pass. Then set the whole thing to run on its own.
Muted · captions on
Video chat
Ask in plain English, jump to the moment that answers it
The answer arrives with the timecodes it came from, and every one of them is a click away.
Muted · captions on
Batch
Clear a hundred clips in the time it takes to watch one
Send the whole upload at once and read the findings side by side when it lands.
Muted · captions on
Schedules
Set it once and the answers keep arriving
A workflow that runs to its own schedule, with every past run kept on the record.
04 / THE SURFACE
One engine. Every read.
Detection, retrieval, reasoning, and reporting run off the same understanding of your footage — so the breadth compounds instead of fragmenting into tools.
DETECT · TRACK
Frame-by-frame analysis
Objects, actions, and events detected and tracked with persistent IDs and timecodes.
Upload, connect Drive, or point at a link. Batch as many clips as you like.
02
Detect & track
Every minute analyzed — objects, actions, events, with timecodes.
03
Index
Embeddings make the whole library searchable in natural language.
04
Ask
Query or converse with the agent; it reasons across clips and cites frames.
05
Report
Synthesize findings into a structured, exportable document.
06 / THE PAYOFF
From a thousand minutes to one document.
When the read is done, synthesize it — an executive summary, findings anchored to timecodes, and the numbers, in a document your team will actually read.
Findings tied to timecodes Every claim points back to the exact second in the source.
Numbers, not vibes Counts, dwell times, and deltas extracted straight from the footage.
Branded & exportable A clean PDF that reads like it came from your team, not a tool.
Q3 Site Read — North Gate
synthesized from 14 clips · 3h 41m
READ #0428 2026-07-28 ● COMPLETE
02:14Loose guardrail, north stairwell — recurs across 4 sitesSAFETY
08:47Two fire-extinguisher tags expiredCOMPLIANCE
21:03Loading-zone dwell up 12% vs Q2OPS
07 / REPEATABLE
Configure intelligence once. Run it forever.
A template is the set of questions you always ask, saved. Point it at next week’s footage and the same read comes back in the same shape — which is what makes two batches comparable.
01
Save the questions
Get one read right — the prompts, the depth, what the report should contain — and keep it instead of rebuilding it next time.
02
Point it at the next batch
New clips, same template. Run it across a whole upload at once rather than clip by clip.
03
Get the same document back
Same structure, same fields, same numbers in the same places — so this month can be laid next to last month and actually read.
Reads teams set up first
TRAFFICTraffic study
How many, how fast, and what went against the signal.
SAFETYWarehouse safety
People near moving equipment, missing PPE, blocked routes.
FLOWRetail flow
Where people go, where they stop, and for how long.
AUDITSecurity audit
Access events, restricted areas, anything out of hours.
08 / QUANTIFIED
The read, as data.
Everything detected becomes a number you can track — throughput, dwell, exceptions — trended over time.
Throughput / hr
0
▲ 6.2% vs last wk
Avg dwell
0s
▲ 12% vs last wk
Exceptions
0
▼ 18% vs last wk
Coverage
0%
▲ 0.3% vs last wk
DEVELOPERS Beta
Your library, from your own code.
The reads you get in the app, available over HTTP: list a library, run an analysis, poll the batch, search in plain language. Scoped keys, one shared credit balance, and an OpenAPI 3.1 spec.
A per-seat platform fee with a generous pooled credit bundle that refreshes every month — credits are spent by the minute of video, at the depth you pick: 1 credit a minute for a fast read, 5 a minute at standard depth. Heavy reads pay for the compute they use; everyone else stays flat.