AI that watches
every camera,
so no one has to.
Ask a question in plain English. Get the answer — and the clip it came from — in seconds.
- 0+
- cameras connected
- 0.0B
- hours indexed
- 0.0s
- median answer time
One person in a grey hi-vis vest, 14:32:06. Same track exits Gate B at 14:33:41 — badge #4521.
Running today in
Founder
Antler India cohort
Our founder came through Antler India's cohort, one of the country's most active early-stage programmes. We are building the video-understanding layer for the hundreds of millions of cameras already deployed — and almost never watched.
The problem
The footage was always there. Finding it was the problem.
The average site records thousands of hours a month and a human watches almost none of it. Cameras stopped being a security system a long time ago — they became an archive nobody has time to open.
- Someone scrubs six hours of tape to find a ninety-second event
- Investigations wait for the one person who knows the camera layout
- Anything older than last week is effectively unsearchable
- Most incidents are never found at all — nobody was watching
Average time to close a footage request
- Ask a question, get the clip and the timestamp back in seconds
- Anyone on the team can investigate — no camera map required
- Every hour ever recorded stays searchable
- Standing questions watch the feeds so nobody has to
Median time from question to cited answer
How it works
Three steps between you and every answer
Connect what you already own
Point us at your NVR, VMS or RTSP streams — Hikvision, Dahua, Milestone, Genetec, Verkada, ONVIF. No rip-and-replace, no new cameras, no truck roll.
Typical first site live in under 30 minutes
We index every frame
People, vehicles, objects, movement, text and time are extracted into a searchable timeline. Historical archives backfill while live feeds keep streaming.
Backfills ~400 hours of archive per hour
Ask in plain language
Type the question you would have asked a guard. Get an answer with the exact cameras, timestamps and clips it came from — ready to export.
Every answer is citation-backed
Three parcels left Dock 3 at 14:32:06, carried by one person in a grey hi-vis vest. The same track exits through Gate B at 14:33:41 — badge #4521 scanned 4 seconds earlier.
Every answer links back to the exact frame it came from.
Technology
The AI under the hood
Not motion detection with a chatbot bolted on. A multimodal video index, built on Gemini and Vertex AI, so that natural language questions resolve against what the camera actually saw.
- 1Ingest
RTSP, ONVIF and archive pulls, normalised and de-duplicated across every recorder on site.
- 2Perceive
Gemini multimodal models on Vertex AI read each frame — people, vehicles, objects, actions, signage and text.
- 3Index
Embeddings and structured attributes are written to a per-site temporal index.
- 4Reason
Questions decompose into retrieval across time, cameras and entities, with tracks re-identified between views.
- 5Ground
The answer is composed only from retrieved frames, each one cited back to its timestamp.
Multimodal, not metadata
Classical VMS analytics count boxes crossing a line. We describe what is actually happening in the frame, which is why open-ended questions work at all.
Cross-camera re-identification
Entities persist across cameras, buildings and hours. That temporal graph is what makes multi-hop questions — follow this person to the exit — answerable.
Retrieval-grounded answers
Every claim traces to retrieved frames. When the footage does not support an answer, the system says so instead of inventing one.
Runs where the data is
Inference is deployed into the customer environment rather than ours, so operators who cannot let footage leave the building still get the full product.
Platform
Built for the questions security teams actually ask
Not another dashboard of motion alerts. A record you can interrogate.
Search across every camera at once
One question sweeps your whole estate. Follow a person or vehicle as they move between cameras, buildings and sites without switching a single feed.
Timeline reconstruction
Ask what happened between 2pm and 4pm and get an ordered narrative — arrivals, handoffs, exits — instead of six hours of footage to watch.
Standing alerts
Turn any question into a watch. Missing PPE, a door propped open, an unscheduled vehicle in the yard — you get told, not the archive.
Evidence you can hand over
Every answer cites its source frames. Export a clip bundle with hashes and an audit trail your insurer, HR team or law enforcement will accept.
Works with your archive
Months of old footage become searchable on day one. Investigations that used to mean scrubbing tapes now take a sentence.
Cloud or on-prem
Run it in our cloud, in your VPC, or fully air-gapped on a box in the server room. Same product, your data boundary.
Results
What changes in the first month
less time scrubbing footage
Measured across 120 investigations
more incidents surfaced per shift
vs. manual review baseline
average time to export evidence
Down from 4h 20m
fewer false alarms escalated
Context-aware review before dispatch
“A pallet went missing on a Friday. Previously that is two people and half a day of tape. We asked one question and had the clip, the badge and the gate exit before the shift ended.”
Operations Lead
National 3PL, 6 distribution centres
“We stopped hiring for footage review. The team now spends its time on the twelve incidents that matter instead of the four hundred that do not.”
Head of Loss Prevention
Retail group, 210 stores
“Our safety audits used to be a sample. Now every PPE breach in the plant is on a list every morning, with the clip attached.”
EHS Manager
Food manufacturing, 3 plants
Security
Your footage stays yours
Video is the most sensitive data most organisations hold. We treat it that way — with a deployment model that lets the footage never leave your network at all.
AES-256 everywhere
Encrypted in transit and at rest, with per-tenant key isolation.
On-prem & air-gapped
Deploy inside your network. Footage never has to leave the building.
SSO, RBAC & audit logs
SAML/OIDC, granular camera-level permissions, and a full query audit trail.
GDPR & CCPA aligned
Retention controls, redaction, and documented data-subject workflows.
Pricing
Priced per camera, not per headache
Every plan includes unlimited users, clip export and your full archive backfill.
Starter
A single site getting its first answers.
- Up to 5 cameras
- 100 queries / month
- 30-day searchable history
- Email support
- Clip export
Professional
Most popularMulti-site teams running real investigations.
- Up to 25 cameras
- Unlimited queries
- 90-day searchable history
- Standing alerts
- Multi-user access & RBAC
- API access
- Priority support
Enterprise
Estate-wide deployments with a data boundary.
- Unlimited cameras & sites
- Unlimited retention
- On-prem or air-gapped deployment
- SSO/SAML & custom integrations
- Dedicated success engineer
- 99.9% uptime SLA
14-day trial on every plan. No card required. Cancel from the dashboard.
FAQ
The questions we get asked first
Something not covered here? .
Do we need to replace our cameras?
No. We sit on top of the cameras and recorders you already run. If it speaks RTSP or ONVIF — or you use Milestone, Genetec, Hikvision, Dahua or Verkada — we connect to it and start indexing.
How far back can we search?
As far back as your recorder keeps footage. We backfill archives at roughly 400 hours per hour of processing, so months of history usually become searchable within the first few days.
Does footage leave our network?
Only if you want it to. Enterprise deployments run entirely inside your VPC or on hardware in your server room, including fully air-gapped installs. In cloud deployments, data is per-tenant isolated and encrypted with AES-256.
How accurate are the answers?
Every answer ships with the frames it was derived from and a confidence score. Nothing is asserted without a citation, so your team verifies in one click rather than trusting a black box.
Can we use this as evidence?
Yes. Exports include the original clip, hashes, camera metadata and a full audit trail of who queried what and when.
How long does setup take?
Most first sites are answering questions the same day. Connecting a recorder takes minutes; the only real wait is the initial archive backfill.
Stop watching footage. Start asking it questions.
Connect one camera and ask your first question today. It takes about as long as this page took to read.