A camera archive you can talk to — built for the Riverside Mill camera system, now a product.
Finding the right footage across hundreds of cameras and months of retention is a large manual task. Surveillance Broker turns the archive into something you can browse on a true wall-clock timeline, search by what actually happened, and simply ask — and it opens the exact camera and time in one step.
The Riverside Mill campus runs over 200 cameras with 90 days of full retention. That coverage is only useful if you can find the moment you need — and on a system that size, running 24/7, that is a genuinely hard problem. We built an intelligence layer over the camera archive so operators could stop scrubbing tape and simply describe what they were looking for.
It worked well enough at Riverside Mill that we generalized it into its own product: Surveillance Broker. It is not tied to any one site or camera vendor. It runs today on the live Riverside Mill archive — thousands of recordings across every camera in the building — and the same pipeline takes any site's cameras next.
Surveillance Broker sits on top of a proven NVR and adds the retrieval, memory, and conversational control that camera systems normally lack.
Surveillance Broker is two cooperating systems — a local system at the site and a cloud system for centralized access — sharing one domain implementation across every interface.
A proven open-source NVR handles live cameras, recording, and review; Surveillance Broker builds the intelligence around it rather than reinventing it. Recordings are decoded once into a clean, standardized archive and, in the same pass, sampled for analysis. Lightweight detection runs first; only when something is actually happening do local vision-language models describe the people, activity, objects, and visible text in the scene. Those observations become searchable records, embeddings, and a knowledge graph — all kept alongside the camera, time, and source frames they came from.
The cloud system receives authenticated archive uploads and catalogs them immutably. It serves the timeline, playback, and export experience, and runs hybrid search — relational, full-text, vector, and knowledge-graph retrieval combined and reranked — behind the conversational agent. Every answer carries stable references that become viewer deep links and export boundaries.
Long-running jobs — import, processing, enrichment, upload, reconciliation — use durable checkpoints and idempotent operations, so nothing is silently lost and a repeated upload is always safe. A single source of truth holds the authoritative catalog; the knowledge graph is a rebuildable projection on top of it. And there is one implementation behind many front doors: the same operations serve the web UI, scheduled jobs, and other authorized agents over a common tool interface.
A system that understands surveillance footage has to be built carefully. These are architecture decisions, not afterthoughts.
Surveillance Broker grew directly out of the camera, network, and security work we delivered at the Riverside Mill Complex — one of several systems on that campus that started as an engagement and became reusable capability.
See the Riverside Mill camera system it grew out of →