Operators spend 2–6 hours scrubbing timelines to find a 12-second clip. Xplug.Vision replaces that with a plain-English query — type what you're looking for, get the exact moment back, inside the console you already use.
A background service does the heavy lifting continuously, so the operator's query is instant — and never leaves your network.
The service taps Security Center video streams and samples keyframes from every camera, around the clock.
Claude's vision model describes each keyframe in natural language — objects, attributes, colours, context.
Descriptions become vector embeddings in a self-hosted database, timestamped and linked to camera + archive.
The operator types a query; ranked, thumbnailed results jump Security Desk straight to that moment.
No filters, no camera-by-camera scrubbing. Describe what you're after — “person in a red jacket loitering by the loading dock after 9 PM” — and get timestamped, thumbnail-previewed results in seconds.
A background service describes keyframes with a vision model and stores them as vector embeddings, so every camera is always searchable — historical and live.
Clicking a result jumps Security Desk directly to that moment in the archive. No external UI, no new workflow — it lives inside the operator's existing workspace.
Built for hundreds of cameras, multiple operators and 24/7 indexing — with a self-hosted vector database and full audit logging. Footage never leaves your infrastructure.
Every component runs inside your network. The pipeline is continuous and asynchronous — indexing never blocks the operator, and queries never wait on it.
Xplug.Vision is in active development. Join the early-access programme to pilot it on your deployment and shape the roadmap.