KragAi is a smart AI unit that
recognises, remembers, decides
according to rules, and answers. Locally, in real time.

A machine that can hold a post.

It stands at an entrance and holds it: it knows the people who belong, speaks with whoever is there, decides only what it has been allowed to decide, and keeps a record of why.

  • All of it on hardware inside your building.
  • Nothing goes to a cloud.
The KRAG Smart Unit: a dark wall-mounted box with a pan-tilt camera head, a microphone button and speaker slots
Concept render
  • Observe
  • Listen
  • Think
  • Speak
  • 30 FPS4K resolution
  • DynamicGPU savings
  • 0%Data leak risk
  • 100%Local inference

Cameras record, and someone watches the footage afterwards.

Analytics send an alert, and someone still has to look, work out what is happening and decide.

Nobody answers for the decision, and there are never enough people to hold every post at every hour.

Who's at the entrance?

The product isn't the detection. It's the decision, the conversation and the record of why.

An illustrative scene. The names are made up; the behaviour is the product's.

An entrance with a KRAG unit on the wall and a person at the post

Hi Lydia. Jack's on his way.

GUARDIAN15:31

Lydia Marom

Collecting Jack, Group B

How sure
Decision
Allowed
Because
Listed guardian

Every decision, accounted for.

When someone new asks to collect Jack, KRAG says no, calls Dana, and writes down why. The same record keeps every routine arrival too, so anyone can check afterwards who came, what was decided and on whose authority.

TUE 15:42 · ENTRANCE ANeeds a person
Who
Unknown adult
How sure
no match
Asked
To collect Jack, Group B
Rule
Only listed guardians collect children
Decision
Refused. Dana Levi called to the entrance
Authority
Pickup list, kept by Dana Levi
  • 15:38Rami Amar, visitor pass until 17:00Escorted
  • 15:31Lydia Marom collected JackAllowed
  • 15:12Noa, staff, back from lunchAllowed

Illustrative record. Names are made up.

Holding a post means four things at once.

Four people framed by corner brackets that change colour from unknown to face and voice confirmed
  1. Know who

    Face, body and voice together, with an honest sense of how sure it is. A weak recognition never carries the weight of a strong one.

  2. Remember

    Who came, who left, who is inside now, what was said and what was decided.

  3. Decide, within limits

    Your building's own rules: who may enter, who may collect whom, who must be escorted. Every rule has an author, and authority can be withdrawn.

  4. Answer for it

    Every decision keeps its evidence: who, what, when, and on whose authority.

It runs inside the building.

Video, audio and identity data stay on your site, because the whole system was built to run there. It also keeps working when the internet connection drops.

Four Smart Units connected to one KRAG Core inside the building, with the cloud crossed out

The AI never decides.

Every decision comes from a rules engine that gives the same answer every time. The language model only chooses the words, and it can't soften a refusal. That is why a talking machine can be trusted at a real post.

Face, body and voice feed a trust policy and a deterministic engine that makes the decision; the language model only supplies words

A new building is a configuration.

The engine doesn't know what a "child" or a "contractor" is. Each building describes its own people, rules and procedures in a mission package, and the engine stays the same.

Mission packages for different buildings plugging into the same engine

What KRAG is not.

Not a camera system.
It works from video, but it doesn't sell storage or hours of footage to search.
Not video analytics.
Analytics stop at an alert for someone to handle. KRAG handles it: it decides, speaks, refuses or calls a person.
Not a badge reader.
A badge proves someone is holding a card. KRAG knows the person, in context, and you can talk to it.
Not surveillance.
It doesn't watch everyone in case something turns up. It looks after one post and the decisions made there.

Small sites where the people who belong are known.

KRAG suits places where the people who belong are a known group, and the decision at the post follows a rule.

  • CHILDCARE

    Who may collect this child today

    And whether any child is left unattended.

  • CARE AND SPECIAL NEEDS

    Nobody leaves unaccompanied

    Staff know the moment someone tries.

  • CLINICS

    Staff and today's patients

    With nothing leaving the premises.

  • DATA CENTRES

    This contractor, today, with an escort

    As the site's rules require.

  • OFFICES AND LOBBIES

    Known people and expected visitors

    And a record of both.

And the posts nobody staffs today. The side entrance. The after-hours lobby. The unattended gate. Once a post can be held for a fraction of a salary, buildings hold posts they never held before.

Two pieces.

A Smart Unit on the wall at the entrance, cabled to the KRAG Core in a technical room

KRAG Core

A compact unit, installed out of sight. All perception, reasoning, language and storage happen here, on site. It holds your building's rules and its record.

KRAG Smart Unit

The visible device at the post: camera, microphone, speaker and a status light, on one Ethernet cable. The Core is designed to hold up to four posts.

See a live kindergarten demo.

KRAG running the kindergarten package at our office: recognising people, granting and refusing entry, delivering messages, and keeping the record. Unscripted; we only cut the waiting.

7:48 · captions available in the player.

Built and running.

KRAG works end to end on live video today: recognition, memory, spoken interaction at the post, rule-based decisions, occupancy, authority and a full audit record. That is around 102,000 lines of code and 3,838 automated tests, built over twelve months by two engineers. The hardware unit is specified, costed and entering build.

KRAG also builds on model research we developed in-house.

Passing tests show the software does what we designed it to do. They don't show how it performs in a real building, so we don't quote accuracy figures yet. Measuring that at the first real sites is the next step.

Not yet, stated plainly

  • No installation in a real building yet
  • No measured false-alarm or miss rate on real traffic
  • IP camera input not wired yet
  • More than one post per Core, and opening locks and barriers itself: designed, not built
KRAG at an entrance, in 22 seconds. Scenes and names are illustrative.

Frequently asked questions.

What is KRAG?

KRAG is a smart AI unit, made by KragAi, that holds a post: a place with a responsibility attached, such as a school entrance at pickup time or a clinic reception. It recognises the people who belong, remembers who came and went, decides according to the building's own rules, and speaks with whoever is there. It keeps a record of why, and runs locally, in real time.

Does any video leave the building?

No. KRAG Core, a compact unit installed inside the building, does all the recognition, reasoning, speech and storage. Video, audio and identity data stay on site because the system was built to run there, not because of a setting, and it keeps working when the internet connection drops.

Does the AI make the decisions?

No. Every decision comes from a rules engine that applies the building's own rules and gives the same answer every time. The language model only chooses the words KRAG says, and it can't soften a refusal. Every rule has an author, and that authority can be withdrawn.

What happens when KRAG isn't sure who someone is?

It doesn't guess. KRAG combines face, body and voice and keeps an honest sense of how sure it is, so a weak recognition never carries the weight of a strong one. When it is unsure, it asks. When something is new, it calls a named person, with the reason attached.

Where does KRAG fit?

In buildings where the people who belong are a known group and the decision at the entrance follows a rule: childcare and schools, care and special-needs facilities, clinics, data centres, and offices and lobbies. It also fits entrances nobody staffs today, such as a side entrance or an after-hours lobby.

What is built today, and what isn't?

KRAG runs end to end on live video today: recognition, memory, spoken interaction at the post, rule-based decisions, occupancy, authority and a full audit record. You can watch it working in our live kindergarten demo. It isn't installed in a real building yet, so we don't quote accuracy figures.

Who we are.

KragAi is an Israeli company building local AI for the places where decisions are made in person.

Isaac Perez

Isaac Perez

CEO and technical founder

isaac@kragai.io

Thirty years in software and systems engineering: embedded Linux, real-time and secure systems. B.Sc. Computer Engineering, Technion.

Yarin Perez

Yarin Perez

CTO

yarin@kragai.io

Embedded systems, computer vision and edge AI: GPU-accelerated vision pipelines and connected devices in medical and defence programmes.

We're raising our seed round.

KRAG already works end to end on live video. The next step is the first real installations, and the measured record of how it performs in them. If you invest at this stage, Isaac would like to show you KRAG running live and walk you through the plan.

Contact Isaac for more info.