Live video
Build agents for live video
Nobody can watch every feed. VideoDB reads cameras, drones and screens as the video arrives, raises the events you describe in plain language, and sends each alert with the clip behind it.
Six feeds on this wall, and hundreds more behind it
“Tell me when someone
walks into the yard”
The alert arrives seconds later, with the clip behind it
Watch for anything you can describe
Connect a feed and write the event as a sentence. VideoDB reads the video in short windows and sends an alert when one of them matches.
Open the notebook (opens in a new tab)- Sources
- RTSP cameras, NVRs and drone gateways, plus screens and microphones, from one feed to thousands
- Events
- A sentence, such as “a person is checking the lock”. Edit it to watch for something else
- Models
- A vision-language model with your prompt, or your own detector as an analyzer
- Alerts
- Over a webhook or a WebSocket, each with the reason it fired and a clip of the moment
- History
- With storage on, every feed stays searchable in plain language, and each result is a clip
Try it on a live stream
Two demos run on sample feeds: a camera on a property, and a dashcam on the road.
The demo plays live video from a sample camera.
Start the live demo (opens in a new tab)Start from an event that already runs
Each row is a notebook on a public sample stream. Change the sentence to watch for something else.
The cookbook on GitHub (opens in a new tab)Notebooks for live video, each with the event it watches for. Each opens in Colab in a new tab.
- Property “Detect if a person is interacting with the door or visibly checking the lock.”
- Roads “Detect if an accident or vehicle collision takes place.”
- Floods “Detect sudden flash floods or water surges.”
- Homes “Detect if the baby is trying to escape or climbing out of the crib.”
- Live sport “Detect when a batsman hits a six.”
- Public spaces “Detect luggage left unattended for 1+ minute.”
Watch a feed in code, or in a sentence
One file connects a camera, describes the event and says where the alert goes. In Claude or ChatGPT the same jobs are plain requests.
import videodb
conn = videodb.connect()
coll = conn.get_collection()
# 1. Connect the feed and keep its history
cam = coll.connect_rtstream(
url="rtsp://samples.rts.videodb.io:8554/intruder",
name="Front door", store=True)
# 2. Read it in 5-second windows and index what the model sees
seen = cam.understand(
segmentation={"type": "time", "window": "5s"},
analyzers=[{"type": "vlm", "name": "scene",
"sampling": {"frame_count": 2},
"config": {"prompt": "Describe the people and what they do."}}],
store=True)
index = cam.index(name="front-door", source=seen.outputs["scene"],
use_for=["semantic"])
# 3. Describe the event, and say where the alert goes
event = conn.create_event(
event_prompt="Detect if a person is checking the lock.",
label="lock_checked")
index.create_alert(event_id=event,
callback_url="https://example.com/hooks/camera")
# Later: ask what happened, and play the answer
found = cam.search(query="a delivery left at the door", index_id=index.id)
found.get_shots()[0].play()
{
"label": "lock_checked",
"confidence": 0.92,
"explanation": "A person stops at the front door and tries the handle.",
"stream_url": "https://….m3u8"
}
- explanation
- Why it fired, in a sentence
- stream_url
- The moment, as a clip that plays
What is the front door camera showing right now?
Tell me when someone checks the lock.
Show me every delivery from this week.
The stream in this file is a public sample, so it runs as it is with a free API key.
From the first feed to a whole site
A four-week pilot on your own cameras, in our cloud or yours.
- Week 1
First feeds live
Ten cameras connected, the events agreed, and an engineer from our team working with yours.
- Week 2
Models and dashboards
Your own model runs as an analyzer where you have one, and alerts reach the tools your operators already use.
- Weeks 3 and 4
The whole site
A hundred feeds or more, with the alerts checked against incidents you already have on record.
Build it into your own product
For companies that sell monitoring, and for the integrators and clouds that serve them.
- Brand
- VideoDB runs inside your product, under your name
- Models
- Detectors you have trained keep running, with the feeds, the index and the alerts around them
- Region
- Deployed in your cloud, in the region your buyers need
Run in our cloud or yours
VideoDB is video infrastructure for agents and devices that see. 1,000+ developers and agents use it every week, and it has processed 300+ TB of video.
SecurityRuns on any cloud
AWS, Google Cloud and Azure, or your own cloud.
AWSGoogle CloudAzure
Security and compliance
SOC 2 Type II, GDPR and ISO 27001 compliant.
Run a pilot on your own cameras
Tell us what you need to catch and where the cameras are. We reply within one business day.