What are QpiAI Vision Agents?

Vision Agents turn a task written in plain language into a complete visual workflow. The agent plans the steps, picks the models for each one, runs them on your images or video, checks the results and hands back outputs your team can use.

How is this different from a single computer vision model?

A model answers one narrow question, such as where the people are. Vision Agents chain detection, segmentation, tracking and vision-language models together, decide which one to use at each step, and verify the result before reporting it.

Can I ask questions about a video?

Yes. After a video is analysed you get a written summary, and you can ask follow-up questions such as which lane had the longest queue. Answers point to the moment in the video they refer to.

Can it follow the same person across cameras?

Yes. Multi-camera analysis re-identifies a person as they move between camera views and places them on a site map, so you can follow one track through a whole site.

Which industries is it built for?

QpiAI Pro serves Healthcare, Finance, Legal, Media & Entertainment, Construction, Space & Environment, Manufacturing, Retail, Transportation, Agriculture, Energy, Defense, Education, Customer Service, HR & Recruiting, Content & Marketing, Math & Reasoning, and Code / SWE.

Vision Agents is in early access. Email pro-support@qpiai.tech or request access at www.qpiai-pro.tech, and our team will set up your workspace.

Beyond detection.Vision agents that see, decide and act.

Describe a goal in plain language. Specialized agents plan the work, analyse the footage and deliver alerts, reports and next steps.

Get Early Access

Alert sent · High priority

Checkout queues over the limit

0 of 4 lanes have more than 6 customers waiting. Lane 2 is the longest at 0.

Sent to: Floor manager
Evidence: 2 verified frames
Recommended: Open another till

Store audit report · Prepared by Vision Agents

Supermarket 2 · Camera 1 · Checkout · 2:11 of footage

PDF Share

0/100Needs attention

771People tracked

5Staff on shift

7Customers billed

1:06Avg time to billing

20Peak queue

87%Lane utilisation

SOP results

FailCheckout queues stay short and moving2 high findings
PassCheckouts staffed and attentive4 of 4 lanes staffed
N/ACheckout area clean and clearNot in camera view

Agent summary

Long queues of roughly 10–16 people per lane persisted for the whole clip. All four lanes were staffed throughout. 7 customers were billed, with billing times from 0:15 to 1:41 (average 1:06); the cashiers were busy for most of the clip and the queues did not shorten.

Findings · verified on evidence

Evidence frame from the checkout footage: Lane 1 bracketed, 11 people waiting against a limit of 6, marked verified. — **Queue length exceeded**Lane 1 · 11 waiting · limit 6

Evidence frame from the checkout footage: all four lanes outlined in red and labelled over limit. — **Long queues in several lanes**4 of 4 lanes over the limit

Floor manager alertedRecommended · open another till at peak

One sentence in.
A verified audit out.

Get Early Access

Everything you need to see and act.
All in one platform.

Turn natural language instructions into structured visual workflows that agents can coordinate and execute.

Diseased tomato leaf detector All systems OK

Describe the outcome and the success criterion. Pro Vision plans and runs the rest.

Compose vision workflows on a canvas: drop models, logic and visualisations from 69 building blocks, connect them and run a preview on your own image.

Workflows › Visual builderCheckout people counter Unsaved changes

Canvas1 blocks0 connectionsDrag bottom circle → top circle

InputWorkflow input

Workflow sourceOutput ↓

Ready to connect

Run preview

Detect, segment and track objects across live or recorded video, then get a summary of what happened and ask questions about it.

Checkout camera 1 · Video intelligence Analysing All systems OK

Detecting · 0 people

0:002:11

Ask about people, queues or events in this video…

Map every camera onto your floor plan, so each person becomes one point on the site with the same ID in every view.

Campus · 2 cameras · floor plan Mapping All systems OK

Campus floor plan. — **Floor plan** campus map

Camera 1 Hospital road

Camera 2 School

Spatial analysis0/4 done

Map cameras to the plan
Track people in each camera
Place them on the floor plan
Link one ID across cameras

On the plan–
Camera 1–
Camera 2–
ID 24 walked–

Coordinate multiple camera feeds to analyze activity, track objects and build a unified understanding of environments.

Eye Glass Store 1 · 4 cameras · 15 Jan 2019 Re-identifying All systems OK

Camera 1 Sales floor · 7 people

Camera 2 Reception desk · 1 person

Camera 3 Waiting area · 3 people

Camera 4 Front of store · 2 people

Re-identification0/3 done · matching 0/4

Track people in 4 camerasSAM 3 tracker
Compare appearanceClothing, colour and build
Link tracks to one identity4 tracks · 1 shopper

Camera 1 · track 3
Camera 2 · track 1
Camera 3 · track 1
Camera 4 · track 2

Set zones and rules once. Vision Agents flag every break the moment it shows on camera, with the evidence attached.

Eye Glass Store 1 · Camera 5 · Office Monitoring All systems OK

Zone · Staff-only area IR night mode

All rules holding3 rules active on this camera · night mode

0:09 · Office window forced
0:14 · Intruder climbing in
0:18 · Intruder in staff-only area

Outcomes

Fewer blind spots.
Faster decisions.

QpiAI Pro turns images, video and live camera streams into results your team can act on.

Manufacturing · Scrap contamination check

01 / 09DetectionTrackingAnomaly flag

FAQ

Questions,
answered.

Something else on your mind? Our team is happy to walk you through it.

Vision Agents turn a task written in plain language into a complete visual workflow. The agent plans the steps, picks the models for each one, runs them on your images or video, checks the results and hands back outputs your team can use.
A model answers one narrow question, such as where the people are. Vision Agents chain detection, segmentation, tracking and vision-language models together, decide which one to use at each step, and verify the result before reporting it.
Yes. After a video is analysed you get a written summary, and you can ask follow-up questions such as which lane had the longest queue. Answers point to the moment in the video they refer to.
Yes. Multi-camera analysis re-identifies a person as they move between camera views and places them on a site map, so you can follow one track through a whole site.
QpiAI Pro serves Healthcare, Finance, Legal, Media & Entertainment, Construction, Space & Environment, Manufacturing, Retail, Transportation, Agriculture, Energy, Defense, Education, Customer Service, HR & Recruiting, Content & Marketing, Math & Reasoning, and Code / SWE.
Vision Agents is in early access. Email pro-support@qpiai.tech or request access at www.qpiai-pro.tech, and our team will set up your workspace.

Multi-camera and floor-plan footage: MEVA dataset, mevadata.org, CC BY 4.0 · Quality inspection footage: VisA dataset, Amazon, CC BY 4.0 · Construction and sports footage: Mixkit (free licence), annotated by QpiAI Pro. · Other footage: Pexels stock. The live-camera clip is a recording played as a camera stream.