[ for the cameras you already own ]
Your cameras see everything.
JaneOS notices.
Detection, recording, search, and an agent you can talk to — running on your own hardware, over your own feeds. Threat is a rule you set, never a guess about a person.
01 / PLATFORM
One pipeline from lens to answer
Ingest → detect → record → search → ask. Every stage runs on hardware you control.
▸ INGEST
Any camera you own
RTSP, ONVIF, RTMP, HLS, WebRTC, USB — if you own the stream, JaneOS speaks it. Discovery finds every camera on the network; a human approves each one before it joins.
▸ DETECT
Rules, not guesses
People, vehicles, packages — filtered through zones, schedules, and rules you draw. One person walking through is one event, not fifty. Severity comes from your rules, never from a hunch about a human.
▸ RECORD + SEARCH
Own your footage, then find anything in it
Continuous recording with retention you set and deletion that actually deletes. Search it like a database: “blue pickup, loading dock, last Tuesday, 2–4 pm.”
▸ ASK
Protection you can talk to
“Who was at the back gate this week?” Two enrolled matches, one UNKNOWN, clips attached. It reports what it saw — and tells you plainly when it doesn’t know.
02 / HOW IT WORKS
Cameras in. Answers out.
Video never has to leave the building — events and answers do.
YOUR CAMERAS
any protocol, your VLAN
ONE BOX
ingest · record · detect
RULES ENGINE
zones · schedules · watchlists
YOU
web · phone · agent
Heavy AI on the box — or on your server. Same software, a config choice, not a rewrite.
03 / PRINCIPLES
Enforced in code, not policy docs
Surveillance software has a trust problem. JaneOS is built so the right thing is the only thing it can do.
Your feeds only
Only cameras you own and operate. No third-party feeds, no public cameras, no data brokers. Ever.
People enroll people
A recognition entry exists only because a human added it, with a written reason. The AI pipeline physically lacks the database permission to create one.
UNKNOWN stays UNKNOWN
No fabricated names, histories, or “intentions.” A stranger is a stranger until a human says otherwise.
Hard tenant isolation
Row-level security on every table. Your data cannot appear in anyone else’s queries — including ours.
Bounded retention
Every byte has an expiry you set, and hard-deletion is verified. Unknown-person data expires fastest of all.
Recognition is a choice
Face recognition ships OFF. Turning it on requires a recorded consent configuration for your site and jurisdiction.
Everything audited
Every enrollment, export, playback, and query lands in an append-only audit log. Even we can’t rewrite it.
04 / DEPLOY
Your hardware. Your call.
Same containers in every topology — deployment is configuration, not a rewrite.
one box at the site — visible from anywhere on earth
[ drag to rotate · ctrl+scroll to zoom ]
▸ ONE BOX
Everything on-site
Ingest, AI, recording, and the agent on a single edge box on your network. Works with the internet cut. Nothing leaves the property.
[ AIR-GAP CAPABLE ]
▸ SPLIT
Edge capture, central AI
Thin boxes at each site; your GPU server does the heavy inference. Cameras stay on their isolated VLAN with no internet egress.
[ MULTI-SITE ]
▸ CONNECTED
Viewable from anywhere
The same site, live on janeos.ai — the box dials out over an encrypted tunnel. Nothing dials in, no ports opened.
[ OUTBOUND ONLY ]
05 / THE AGENT
Ask Jane. Get the truth.
Grounded in your footage and your rules — it reports what it observed, never who it guesses.
» did the delivery guy come today?
No enrolled match today. One UNKNOWN person carrying a package appeared at the front door, 11:42 — clip attached.
[ CLIP 11:42:07 · CAM 01 ]
» wake me if anyone is on the north side after midnight.
Rule armed: north fence · person · 00:00–06:00. One alert per event, clip included.
[ RULE #14 · AUDITED ]
» show me every silver sedan in the lot this week.
9 appearances across 3 cameras. Newest first.
[ 9 CLIPS · ATTRIBUTE SEARCH ]
honest by design — “UNKNOWN” is an answer it is allowed to give
[ EARLY ACCESS ]
Put your cameras to work.
Pilot deployments start with a single box and the cameras you already have. Detection first — recognition only if, when, and where you choose it.