Casebound — product walkthrough

The problem we remove

When a child goes missing or a suspect runs, the first minutes decide the outcome — and today those minutes are spent knocking on doors: Do you have a camera? Can we see the last hour? Is the owner even home? Private footage is overwritten in days. Owners are unreachable. By the time a clip reaches an investigator, the trail is cold.

Casebound turns that canvass into one action: police draw an area, every owner inside gets the request on their phone, and a tap grants a time-boxed, single-case, audited window. Where the camera supports it, the feed goes live — and the route assembles itself while the search is still running.

1 tapreplaces a door-to-door canvass
30 minhard cap on any live window
1 casescope of every grant — never standing
0 facesvisible except the person sought

The camera owner joins

A resident puts one camera on the map — and is rewarded for it.

01
Two doors, one product

A citizen with a camera and a detective with a case enter the same system from opposite ends. Nothing here is a general surveillance console.

02
Sign in with Google

Onboarding a camera owner has to be as easy as any consumer app, or nobody joins. One tap, or a demo account.

03
Private by default

The resting state of every camera: no requests, no access, nothing shared. Police see nothing until this person says yes.

04
A name and a pin

The whole onboarding: name the camera, drop it on the map. The pin is what lets a detective find the right cameras later — without a database of anyone’s address.

05
On the map, still private

The camera is registered and marked PRIVATE. It only ever streams for one approved case.

06
Verify your own camera

The owner previews their real Cam720 feed to confirm it works — before granting anyone anything. This is a live RTSP camera going through the deployed streaming fabric.

07
Rewarded for coverage

Connecting a camera credits the owner immediately — a ledger line, not a promise. Contribution to a closed case pays more on top.

08
Redeemable for real civic perks

K9 unit meet & greet, a patrol ride-along, a VIP spot in the Stampede Parade. This is why owners opt in — and stay in.

The call comes in

A missing-person report. The clock starts.

09
The other end of the network

A Calgary Police detective signs in to the same product the camera owner just joined.

10
A case is bounded from second one

A named authorization, a case type, and a mandatory reference photo of the person being sought. A case cannot be opened without one.

11
Timeframe, sharing, signature

A fixed footage window, an authorization lifetime, and a signed attestation under lawful authority. The photo can be shared with owners like a missing-person poster — or kept police-only.

12
Draw the canvass on a map

Every camera in the area, in one view. This replaces knocking on doors while the footage is being overwritten.

Consent in seconds, not days

The request reaches the phone before an officer could reach the street.

13
The request lands on the phone

The missing-person photo, the case, the authorization reference, the exact footage window. No phone tag, no absent owners, no door-to-door.

14
One tap — signed on the device

The owner can question the detective first. Approval is a cryptographic act, scoped to one case and one window; the server refuses any grant whose signature does not verify.

15
Everyone blurred except the person sought

Not a checkbox — the rule. Bystanders are covered with an opaque mosaic; only the sought person is visible. Revealing any other face is a logged disclosure.

16
Live needs a second, fresher yes

A live look is never standing. The owner allows this specific window, capped at 30 minutes, killable at any second.

17
Live, time-boxed, logged

A real camera streaming into an authorized case, with the countdown visible. The minute that used to be lost to logistics is now spent watching.

18
The route assembles itself

The payoff: not a clip you request and wait for. As the person is seen on successive cameras, the map traces where they went, in order.

19
Every step on the record

Requested, approved, granted, streamed, viewed, matched — an append-only trail. Access ends when the case does, and the data is purged.

Distribution — how cameras actually get connected

Coverage is not a technology problem. Every route below either works today, is waiting on a vendor, or will never exist — and the product says which, in the owner's own portal.

Residents

a doorbell or two

Will never install software. Zero-install paths only.

Clip upload · phone-as-camera · Ring once certified

1–2 cameras per sign-up

Businesses with existing CCTV

shops, gas stations, parkades, strip malls

Already run an always-on machine and already own the problem.

One site bridge, on hardware they already have

4–32 cameras per sign-up

Route in	State	Reality
Upload a clip Every camera ever made — Ring, Nest, Arlo, Wyze, Blink, any NVR, any phone recording	Available now	When police request a window, you export that clip from your own camera app and open the link. Faces are read in your browser; the video itself never leaves your device.
Dashcam and body-worn footage Anything that records to a file — dashcams, action cameras, a phone recording of a screen	Available now	Not a camera connection at all: the owner uploads the clip like any other recording.
Use a phone as a camera Any phone — no camera hardware needed at all	Available now	Point a spare phone out of a window and it becomes a live, consented camera for one case window. Nothing to install.
Site bridge (RTSP + ONVIF) The same cameras as above, plus whole NVRs — one install covers every camera on the site	In development	In development. It is software for a machine the site already has — no hardware is shipped and nothing is given away.
Camera pushes to us (RTMP) The minority of models with native push — some Dahua, some Hikvision cameras not behind an NVR	In development	Planned, but lower priority than the bridge: most camera firmware has no RTMP push, and the biggest vendor does not support it behind an NVR at all.
Business cloud VMS (Verkada, Rhombus, Eagle Eye) Commercial sites already on a cloud video platform — offices, schools, retail chains	In development	Not built yet, but nothing external blocks it: unlike Ring and Nest, these platforms let a customer generate their own API key, so no vendor certification is involved. It needs one adapter per platform.
On-site VMS (UniFi Protect, Synology, Milestone) Sites running their own recording software rather than a plain NVR	In development	Reached by the site bridge with a thin adapter per platform. Synology in particular documents a recording-export call that returns exactly the authorized window, which is better than pulling a live stream and waiting.
Ring Ring doorbells and cameras — live view and past clips	Awaiting vendor approval	Ring opened a public Partner API in January 2026, and it does everything we need. It cannot be switched on until Amazon certifies our application — a vendor approval, not code we can write. Until then, clip upload works for every Ring camera.
Google Nest Nest cameras and doorbells — live view only	Awaiting vendor approval	Needs a Google Device Access project to be approved before it can be enabled. Note the permanent limit even then: Nest exposes no way to fetch past recordings, so older footage will always come through clip upload.
Arlo, Wyze, Blink, Apple HomeKit These cameras specifically	Not possible — ever	These makers publish no usable interface — Arlo has none, Wyze withdrew its RTSP firmware, Blink offers nothing, and Apple clips are end-to-end encrypted. No product can connect them live. Upload a clip instead: it works with all of them, and it is the only path that ever will.
IP camera, direct (advanced) Any RTSP camera or NVR — Hikvision, Dahua, Axis, Reolink, Amcrest, Bosch, Tapo, Lorex, Swann, Annke	Available now	Publish a live RTSP camera straight into an authorized window from a computer on the same network. Works today, but it needs a command or OBS — which is why the site bridge above exists.

Shown to owners in the app, not just in this deck. Ring and Nest stay switched off until each vendor certifies the application — a business gate, not code — and Arlo, Wyze, Blink and HomeKit publish nothing to connect to, so clip upload is their permanent path.

The full-screen product

The owner uses a phone. The detective uses a desk — these are the same build, captured at full desktop width.

D01
The detective’s desk

Every case the officer is authorized on, with live counts of cameras, pending requests and activity. Cases are opened and closed here — there is no standing view of the city.

D02
One case, one page

Cameras granted, requests pending, frames indexed, sightings. The map shows granted cameras in blue, pending as dashed, everything else grey — and the amber pins trace where the person was seen, in order. Every action on the right is an audit line as it happens.

D03
The canvass, as an inbox

Requests to camera owners and their answers, in one thread per camera. This is what replaces knocking on doors while the footage is being overwritten.

What is real, and what is staged

Real and running: the application, accounts and roles, the map, case creation with signed attestation, the area canvass, the request → consent → grant pipeline, the points ledger and rewards shop, the append-only audit — and a real IP camera streaming live through the deployed WebRTC fabric (SFU + TURN relay) into an authorized case.
Demonstration data, clearly labelled in the product: the sought-person persona is a synthetic face — never a real person cast as a suspect or a missing child. The bystanders in the reviewed frame come from a benchmark photo and are redacted by the product itself. The sightings that draw the movement route are seeded: a real person cannot be made to walk past a test camera on command. The redaction and the route are the real shipped features, shown against demonstration footage.
Not yet done: face matching still runs in the browser rather than server-side; demographic-bias certification, threshold calibration on representative footage, a privacy impact assessment and legal sign-off are all still open, and are listed as such in the project’s own tracking documents.
A limit we measured on ourselves: redaction masks only the faces the detector finds, and on real street footage at the shipped detector settings it finds a fraction of them — 1 face across 16 frames where the same model at a larger input size found 8. Faces in that footage were small and oblique, which is exactly what mounted CCTV looks like. So the “every other face is blurred” guarantee holds for the frames shown here and does not yet hold at arbitrary camera distances. It is a detection-recall problem with a known fix (larger input, or person-detection first), it is reproducible with a tool in the repo, and it is a blocker on any real deployment.

The problem we remove

Two doors, one product

Sign in with Google

Private by default

A name and a pin

On the map, still private

Verify your own camera

Rewarded for coverage

Redeemable for real civic perks

The other end of the network

A case is bounded from second one

Timeframe, sharing, signature

Draw the canvass on a map

The request lands on the phone

One tap — signed on the device

Everyone blurred except the person sought

Live needs a second, fresher yes

Live, time-boxed, logged

The route assembles itself

Every step on the record

Distribution — how cameras actually get connected

Residents

Businesses with existing CCTV

The detective’s desk

One case, one page

The canvass, as an inbox

What is real, and what is staged