Talk to your video the way you talk to an assistant.
EdgeTrace exists so that people can reliably question huge video datasets, after the fact and in real time, the way we already question documents and search code today. Every camera, searchable in plain language, on your own network.
Two moments we couldn’t let go of.
EdgeTrace did not start as a product idea. It started as frustration with a system that has the footage and still can’t answer the question.
A car broken into in San Francisco’s Marina. The response was not a search of the cameras that almost certainly saw it. It was advice to go to the Tenderloin and try to buy the belongings back.
The video existed. Nobody could ask it anything. That is not a policing problem. It is a tooling problem.
Two analysts splitting a weekend of footage, one starting Friday night, one starting Sunday, watching toward the middle for eleven hours to identify a single person who had graffitied the police station.
Eleven hours of trained people scrubbing video to answer one question EdgeTrace now answers in seconds.
A decade from now, talking to a million hours of video is as ordinary as grepping a million lines of code.
We already expect to search our documents, our messages, and our code instantly, where every word matters and nothing is out of reach. Video is the one enormous dataset still locked behind manual review.
EdgeTrace makes that footage answer to a question. When it works the way it should, the hours an investigator loses to scrubbing come back as hours spent on the case, and the camera that saw what happened can finally say so, in plain language, in seconds.
The manifesto.
Six convictions that decide what we build, what we refuse to build, and who we build it for.
Video should answer to a question, not a shift.
Today we talk to our documents and our code. We grep a million lines in a second and search a decade of email in the time it takes to read this sentence. Video is the largest dataset any agency holds, and it is the one we still review by hand, one frame at a time. That is the gap EdgeTrace closes: ask the footage a question in plain language, after the fact or in real time, and get the answer back in seconds.
The work is meaningful, so the tool has to be built for it.
Public safety is one of the most consequential things a person can spend their work on. The people doing it deserve software built for their reality, not enterprise tools bent sideways to fit. We design for the analyst tag-teaming a weekend of footage, the detective whose lead is going cold, the operator who has to be right while patrol is still rolling.
Your footage stays yours.
The easy way to build this is to pull every agency’s video into someone else’s cloud and monetize it. We refuse that on purpose. EdgeTrace runs on-prem and air-gapped, on hardware you own, inside your network. We do not train on your data, retain it, or re-purpose it: contractually and technically. Sovereignty is not a feature we added. It is the shape of the product.
Power over video demands a record of who used it.
Making every camera searchable in plain language is a serious capability, and capability without accountability is how this technology goes wrong. So we built the guardrails in first: role-bounded access, per-camera scoping, automatic redaction on export, and a signed chain of custody on every query, view, and export. The same record that holds up in court is what keeps the tool honest.
Every city deserves best-in-class AI of its own.
We believe every city should run its own best-in-class multimodal AI, on its own infrastructure, in service of a superior experience for the taxpayers who fund it. Modern intelligence should not be reserved for the agencies that can afford to send their data away. It should run where the work happens, on the cameras and recorders a department already owns.
Build for the people the giants walked away from.
Plenty of technology companies have decided that public safety, government, and defense are not worth building for. We made the opposite decision. This is exactly the work we want to do, with the people doing the hardest version of it, on the hardware they already run.
Defending a line you can’t afford to let anything cross.
EdgeTrace was named by a former Division I goalkeeper at Stanford. A goalie’s whole job is the line: read the play early, hold your composure, and keep the one thing that matters from getting behind you. Let one in late and the whole team’s fate turns on it.
That is the work, moved off the field. The edge is the line a community is counting on someone to hold. The trace is the thread you follow to hold it: across every camera, in seconds, in time to matter. Instead of keeping a ball out of a net, the save is keeping a dangerous person off the street, or stopping a train before someone steps in front of it.
We are an early team building patiently and on purpose, deployed today with departments like the City of Pleasanton and Pleasanton Police. If the work on this page is the work you want to be doing, come talk to us.
See EdgeTrace running against your camera fleet.
A 30-minute briefing. We bring the demo dataset. You bring the questions your detectives actually ask.