01 / ROBOTICS
Eyes for your next machine.
Bring visual observations into inspection and robotics workflows. Your controller owns movement, decisions and safety.
Build a visual workflow ↗A VISUAL LAYER FOR THE WORLD YOU BUILDGive your machines
Give your machines
a way to see.
Robots. Drones. Connected homes. Turn images and sampled video into observations your software can use—with one API for applications and AI agents.
OpenCV measurementsOptional AI contextAPI + MCP
DOORBELL CAMERA / RECORDED ANALYSISTry your clip ↓
Precomputed AutoVision results on this supplied demo clip. Boxes appear only at sampled times, not continuous tracking. Dashed red: advisory AI grounding. Green: OpenCV detector output. No identity, intent or threat determination.
What will you give vision to?
02 / DRONES
A new perspective.
Send sampled aerial frames into monitoring and inspection workflows. Bring your capture adapter; AutoVision does not pilot aircraft.
Explore frame analysis ↗03 / CONNECTED HOMES
More context. Less guesswork.
Turn authorized camera images into useful event evidence. SmartDetector brings that visual layer into a personalized device and event console.
Meet SmartDetector ↗Illustrative applications, not certified autonomous navigation or life-safety systems. Device connectivity and physical actions belong to your application.
THE LIVE LAB
OpenCV + Amazon BedrockGive your app a pair of eyes.
Try your own image or short video. No signup. No clip storage.
Follow-ups use additional public-demo allowance. Notifications and actions remain with your app. Recorded clips below replay cached reports without AI calls.
With AI enabled, sampled frames are sent to Amazon Bedrock. Use media you have permission to process. AutoVision does not persist uploads.
YOUR SIGNAL STARTS HERE
Pixels → observations → possibilities.
Explore motion across time, OpenCV object detections, and an AI scene summary. Then take the JSON straight to your application.
EXPLORE RECORDED VIDEO ANALYSES
DOG AT NIGHT / RECORDED ANALYSISTry your clip ↓
Inspect pet activity in the supplied camera clip.
Precomputed AutoVision results on this supplied demo clip. Boxes appear only at sampled times, not continuous tracking. Dashed red: advisory AI grounding. Green: OpenCV detector output. No identity, intent or threat determination.
BACKYARD VISITOR / RECORDED ANALYSISTry your clip ↓
Inspect person observations across a backyard camera clip.
Precomputed AutoVision results on this supplied demo clip. Boxes appear only at sampled times, not continuous tracking. Dashed red: advisory AI grounding. Green: OpenCV detector output. No identity, intent or threat determination.
PACKAGE DELIVERY / RECORDED ANALYSISTry your clip ↓
Review delivery activity and visible parcels.
Precomputed AutoVision results on this supplied demo clip. Boxes appear only at sampled times, not continuous tracking. Dashed red: advisory AI grounding. Green: OpenCV detector output. No identity, intent or threat determination.
Supplied demonstration clips with cached AutoVision results, not illustrated animations or live inference. View analysis to inspect sampled images and actual returned boxes. Private uploads above are still analyzed on demand.
Short clips are sampled to at most 12 frames in your browser. Original video and audio never leave the browser. Sampling can miss events. Object detection scores are not calibrated probabilities; AI descriptions are advisory.
REAL EXECUTION / RECORDED REPLAY
Explore the pipeline ↗Watch evidence change the next step.
OpenCV → Decider → specialized lens → evidence → review. Captured with real OpenCV 5 and Amazon Nova Lite; playback does not invoke AI.
Stage 1 of 4 · Paused
Original YOLOX person box
Original evidence stays visible. A missed detection in a crop does not erase it. One supplied frame—not a temporal tracking demonstration.
- Runtime
- OpenCV 5.0.0
- AI model
- us.amazon.nova-lite-v1:0
- Follow-up budget
- 2 tools, then finish
OpenCV 5 observes
The original frame contains one person detection. The Decider receives measured evidence, not an unrestricted instruction.
Actual recorded tool inputs and evidence
{
"person": 1,
"package": 0
}Failure handling and evaluation notes
- Crop re-detection missed a person observed in the full image; initial evidence was preserved.
- The planner's package-uncertainty wording is not directly supported by the returned package detections. Reasons remain advisory and require evaluation.
This smoke test demonstrates real tool selection and evidence feedback, not task accuracy. Human review is advisory; no notification, security action or physical action was performed. Replay timing is presentation pacing; tool durations are available in the recorded evidence.
Only concise decision summaries and observable tool results are displayed—not hidden model reasoning. Recorded 2026-10-14T12:42:40.117Z. Uploaded images and videos use the live streaming pipeline with optional autonomous inspection and bounded demo allowance. Recorded playback makes no AI calls.
From pixels to possibilities.
Useful visual evidence.
Motion measurements, OpenCV object boxes for people, pets, vehicles and other COCO categories, and timestamps. Add optional Bedrock scene context and advisory package grounding, kept distinct from measured results.
See the response contract ↗Your application. Your logic.
Use organization-scoped keys for images and timed frame batches. Stateless face measurements leave consent, names and matching in your application.
See the integration pattern ↗Give agents a visual tool.
Bring image analysis into a larger workflow through MCP. Your agent uses the evidence to choose its next step, with the permissions and human controls you provide.
Connect an agent ↗THE ROAD AHEAD / PLANNED
A home that sees together.
Our home-network vision brings cameras, household devices and robots into a shared observation layer. AutoVision provides perception. SmartDetector brings devices and events together. Future integrations could turn those observations into coordinated, permission-controlled workflows.
Roadmap, not an available home mesh, vehicle network or autonomous control system.
CamerasDevicesRobotsAutoVisionApplications + human control
START SMALL. BUILD SOMETHING BIGGER.
Your vision. Your deployment.
Use the hosted API or prepare a private deployment in your AWS account. Private-install infrastructure and release tooling are built; the AutoVision Marketplace listing is not yet live.