SyncVelo Vision AI
Bring fixed cameras from any maker onto one console, site by site. Watch any of them live in your browser, record them continuously, on a schedule or around alerts, and play back or export what they recorded as evidence anyone can check. Camera passwords stay sealed on the server and every view is logged. AI analytics are on the roadmap: every feature on this page shows its status.
from pressing Watch to a live picture, over WebRTC
makers with stream paths filled in, plus any RTSP camera
life of a stream token, for one person and one camera
camera passwords ever sent to a browser
Entrances, atriums and car parks on one wall.
Each building a site, every gate in reach.
Junctions and crossings, live in the browser.
Docks and yards for the people who run them.
Your own cameras on your phone, passwords kept safe.
Sites, cameras from any maker, live view, media servers, roles and audit: everything in this part works now.
How live view works
SyncVelo decides who may watch before a single frame moves. The video then travels from the camera through the media server to your browser over WebRTC: no plug-in, nothing to install.
SyncVelo checks who you are, that your organization uses Vision AI and that you may see this camera.
The sealed password is opened on the server only, to point the media server at the camera. A camera is pulled only while someone watches.
Your browser gets a token that lasts one minute and opens this camera for you alone. A changed or expired token opens nothing.
Video plays over WebRTC in about a second, and the audit trail records who watched which camera, and when.
Cameras from any maker
AvailableGive a camera's RTSP stream address, or an ONVIF Profile S camera's, and it joins its site. For eleven makers the stream path is filled in for you, channel by channel.
Media servers
AvailableSyncVelo's own media server is installed by the deploy kit and used by default. Prefer to keep video on your own network? Connect your media server, camera by camera: SyncVelo arranges each session and the video flows from your server to your viewers.
Record each camera continuously, on a schedule or around alerts, play back any moment in the browser and export evidence anyone can check, with privacy masks and a record of purpose, retention and access for every camera. Everything in this part works now.
Recording and playback
AvailableChoose how each camera records: continuously, on a weekly schedule in the site's time zone, or around alerts, from up to 10 minutes before a vehicle's tracker alert or an IoT OS device's alarm to up to 30 minutes after. Recordings are encrypted on the server and kept in private object storage, not on the media server's disk. Pick a moment on the camera's timeline and it plays in the browser, the first picture well within two seconds.
Privacy by design
AvailableDraw privacy masks over the live picture to cover a neighbour's window or a public pavement. What they cover is hidden in black in live view and playback and burned into masked exports, while recordings are kept as the camera sent them: masking what is stored comes later with the edge box, and AI analytics, when they arrive, will apply the masks first. Every camera that records also carries its purpose, its retention period and a log of who watched what.
Designed and scheduled, not built. Each analytics type will ship only after it passes its evaluation, with its precision and recall published beside it.
Camera analytics
RoadmapStandard analytics will detect and track people, vehicles and bikes, count who crosses a line, flag intrusion and loitering, and map how busy a place gets, without identifying anyone. Configured analytics will follow, set up and evaluated site by site.
Search and assistant
RoadmapNatural-language search will find clips from a description, using SyncVelo's own image and text models, and every search will be audited. Copilot will answer questions about camera events with the clips that support each answer.
Edge box
RoadmapThe edge box will record and run AI on site, keep working when the link drops and catch up when it returns. It will open only an outbound tunnel, so no inbound port is opened at your site, and it will be managed as an IoT OS device.
Security, privacy and the law
What protects your cameras and recordings today, and what SyncVelo will never build.
Each camera's password is sealed with AES-256-GCM and bound to that camera. No response ever carries it, not even to the people who manage cameras.
A stream token is made only for someone allowed to see the camera. The browser cannot read it, and a token for one camera never opens another.
Another organization cannot see, change or watch your cameras: its request is refused before any token is made.
Viewers watch; managers add, change and remove cameras and media servers. Permissions are checked on every request.
Every camera and media server change is audited, and each camera's access log shows who watched live, played back or downloaded an export, and when.
Its control interface answers only SyncVelo's own server, and it is never pointed at a camera on a private or internal address unless the installation allows it.
Each recording is encrypted on the server with AES-256-GCM and kept in the installation's private object storage. Browsers never reach that storage: playback comes from the console itself.
An export holds the original segments as recorded and a manifest of SHA-256 hashes chained over them, signed by the installation with Ed25519. Check it in the console or with sha256sum and openssl: a changed byte shows.
Privacy masks hide what is not yours to watch in live view, playback and masked exports. Every camera that records has a purpose on record, and its footage is deleted when its retention ends.
What we will never build
Left out on purpose, not postponed.
Number plates and face matching
RoadmapPlate recognition for access control and parking, and face matching for access control with enrolled, consenting people, are planned only as regulated modules: after legal review, per country, where the law allows. Neither is available today.
Every feature
Every feature with its status from our product plan. A label changes only when the feature is built.
Group cameras by site: a mall, a campus, a gate or a home, each with an address and time zone.
Any camera or recorder by its RTSP stream address, ONVIF Profile S cameras included.
Eleven makers' stream paths filled in channel by channel, each marked verified once tested on that maker's hardware.
Sent once, sealed and bound to the camera: never returned, never sent to a browser.
Live in the browser in about a second, on desktop and phone, with no plug-in.
Every camera of a site on one wall, each opened only while someone watches.
One minute, one person, one camera, checked against the organization on every live view.
Installed by the deploy kit and on by default, kept private behind SyncVelo's own server.
Any WHEP-capable server, camera by camera, for cameras on your own network.
Viewers watch; managers add, change and remove cameras. Checked on every request.
Sites, cameras and media servers created, changed and removed, kept in the audit trail.
Who watched which camera, and when.
Vision AI's own console address and sign-in, on the same SyncVelo account as your other products.
Each camera records around the clock, on a weekly schedule in the site's time zone, or around a vehicle's tracker alert or an IoT OS device's alarm, from up to 10 minutes before to up to 30 minutes after.
Sealed on the server with AES-256-GCM and kept in the installation's private object storage, not on the media server's disk. Browsers never reach it.
A timeline per camera for the last day, and playback in the browser from any recorded moment, the first picture well within two seconds.
The original segments as recorded, with a SHA-256 chain signed by the installation. Verify it in the console or with sha256sum and openssl; who exported, when and why is logged.
1 to 365 days per camera, deleted on time, with the storage it needs shown: about 21.6 GB a day for a 1080p H.265 camera at 2 Mbit/s.
Up to 16 rectangles per camera, drawn over the live picture; what they cover is hidden in black in live view and playback. Recordings are kept as the camera sent them.
A clip of the period you choose, re-encoded with the camera's privacy masks burned in.
Every camera that records has its purpose and its retention period on record.
Who watched live, played back or downloaded an export, and when.
In the Vision AI console, starting from a ready-made template: recorded time, coverage and storage today, coverage by camera and the cameras with the most gaps.
In the Vision AI console's Reports section: each camera's recorded time, coverage, gaps and storage by day, and its access log, as CSV, XLSX or PDF.
Will find cameras on the network and read their stream addresses over ONVIF.
Pan, tilt and zoom from the console are planned, with saved positions.
Motion and tamper events reported by the camera itself are planned.
Cameras behind NAT will be able to send their stream out to SyncVelo.
Detection and tracking are planned, without identifying anyone.
Will alert when a line is crossed, a zone is entered or someone lingers.
In and out counts per line, and how full a place is, are planned.
Planned as an early warning beside certified fire alarms, never instead of them.
Will show how busy each zone gets, and where people gather.
Will flag a covered, blurred or moved camera, signal loss and recording gaps.
Helmets and vests, and restricted zones near machines, are planned.
Will find clips from a plain-words description, using SyncVelo's own models. Every search will be audited.
Will give summaries, trends and answers, each with the clips that support it.
Density over a safe limit, counter-flow, sudden running: planned as risk signals for staff to act on.
Planned for observable actions, such as a fall or a raised hand. Never emotions or intent.
Wrong-way driving, illegal stops, red-light crossing and estimated speed on calibrated views are planned. Not evidence for fines.
Will forecast crowd or fire risk and unusual activity for the hour: about places, never about a person.
A box on site will record and run analytics next to the cameras.
Will connect out, so no inbound port is opened at your site.
Will keep working when the link drops and catch up when it returns.
Firmware and models will update the way every IoT OS device does.
For access control, parking and lists you are entitled to keep. Only after legal review, per country, where the law allows.
Enrolled, consenting people only. Only after legal review, per country, where the law allows.
One account
Vision AI has its own console address and sign-in, on the same account as your other SyncVelo products. Sign in once and move between them.
Any IP camera or recorder that streams over RTSP, which is most of them, and ONVIF Profile S cameras by their stream address. For eleven makers the stream path is filled in for you, and each is marked verified once tested on that maker's hardware.
No. Cameras reachable from the internet play through SyncVelo's media server. Cameras on a private network work today through a media server you run on that network. An edge box for sites behind a firewall is on the roadmap.
Yes. Each camera records continuously, on a weekly schedule in the site's time zone, or around alerts: from up to 10 minutes before a vehicle's tracker alert or an IoT OS device's alarm to up to 30 minutes after. Recordings are encrypted in private storage and play back in the browser from a timeline per camera. Exports carry the original segments with a signed SHA-256 chain, or come as a clip with the privacy masks burned in.
None yet: analytics are on the roadmap. Standard analytics come first, such as people, vehicle and bike detection, line crossing, counting, heat maps, and smoke and flame as an early warning beside certified fire alarms. Each type will ship only after it passes its evaluation, with its precision and recall published.
Not today. Plate recognition and face matching for access control with consenting people are planned only as regulated modules, after legal review, per country, where the law allows. Emotion recognition, predicting what a person will do, inferring sensitive traits and scraped face databases will never be built.
Live video travels from the media server to the viewer's browser over WebRTC. With your own media server, it goes from your server to your viewers and SyncVelo only arranges each session. Recordings are encrypted on the server and kept in the installation's private object storage, any S3-compatible store; browsers never reach that storage, and playback comes from the console itself. Camera passwords stay sealed on the server and never reach a browser.
Vision AI is its own product on your SyncVelo account, with its own console and sign-in. Add it from the marketplace in your console where it is available, or talk to us. One sign-in opens all your products.
Get started
Book a demo to see live view, recording, playback and evidence export at work today, and to talk through analytics on the roadmap.
SyncVelo AI
Ask about products, devices, pricing or integrations