Natural language video search
Argu AI turns hours of camera footage into an AI video search you run in plain English, no timelines to scrub and no pre labeling required.
The difference from a keyword tagged archive is that nothing has to be labeled ahead of time. A search for a behavior, a piece of clothing, or an event nobody anticipated still works, because the system reads the footage itself instead of matching tags set up months earlier. One investigator search covered six hours of footage across fourteen cameras in under eight seconds, the kind of result that separates real video search software from a folder of clips waiting to be scrubbed by hand. City operators use the same query interface to review incidents across a smart city surveillance network.
- A single written query runs across every connected camera at once, live or archived, over existing RTSP or ONVIF Profile S feeds at 1080p–4K, 15–30 fps.
- One investigator query covered six hours of footage across fourteen cameras in under eight seconds.
- No pre-labeling, tagging or model training is needed before a question can be asked.
- Footage can be retained on an operator-set 30–90 day window, with GDPR Article 15 governing access requests to what was searched or retained.
- Search results can be exported as a document and sent by email, with redaction available before export.
Homeland-security and critical-facility teams in HLS environments use the same written-query search for incident review, and can apply video redaction before a result leaves the building. The underlying approach is the same one covered in vision agents.
| Axis | Conventional tagged archive | Argu vision-agent search |
|---|---|---|
| What triggers a result | An exact tag match set up in advance | A plain-English description of the behavior, object or event, matched directly against the footage |
| What a human must watch | Every timeline, scrubbed camera by camera | A returned table of matches across every connected camera at once |
| Pre-labeling required | Tags, categories or a trained model built before the question can be asked | None — six hours across fourteen cameras returned a result in under eight seconds |
| Camera / feed compatibility | Requires a VMS with a matching tag schema | Existing RTSP or ONVIF Profile S IP cameras and NVR feeds, 1080p–4K at 15–30 fps, live or archived |
| What gets retained | Tagged clips matching pre-set categories only | Full searchable footage on an operator-set 30–90 day window, with GDPR Article 15 access-request support |
| Export path | Manual clip stitching | Result exported as a document and sent by email in the same session |
