SAAI · Spatial Anonymized Agentic Intelligence

Perfect your space. SAAI anonymized spatial AI by DEEPING SOURCE

Does the AI you use know your space? SAAI knows that space.

Request a consultation See how my space reads

Uses the cameras already installed · no identity data retained

100+
Customer spaces
14Mhours
Hours analyzed
103
Patents held

Olive Young · 7-Eleven · Don Quijote · MARUHAN and 100+ other spaces

01 · Raw
Everything in the space is already recorded. It just is not used.
The cameras overhead already see all of it. The footage only gets pulled up after something has gone wrong.
That rarely happens. Data that is only recorded and never used is worth nothing.
From 2025 consultations with 10 customer sites
02 · Anonymization
To use it, we first erase who it was.
In 2018, DEEPING SOURCE started with exactly this.
Video carries privacy regulation with it. Having a person sit and review it is a burden in itself.
So we erase identity at capture. Raw footage is not stored, and only the shape of behavior remains.
SEAL · no raw footage kept · no human viewing · no re-identification
03 · Spatial intelligence
SAAI does not read one camera at a time. It links them and reads the space.
Conventional video AI only sees inside one camera frame. When the view changes it cannot link the same person back up.
Because identity was erased first, every camera can be linked. One person’s journey through the space holds together end to end.
MTMC · 3D coordinates · re-linking across cameras
04 · Live detection
Know the floor right now without going there.
Stockouts, cleanliness, chiller temperature and anomalies are caught while they are happening.
One site or many, processed at the same time. What a camera cannot see, like chiller temperature, comes from a separate sensor read alongside it.
Ready meals 52s · Drinks 38s · Snacks 29s · Magazines 14s
Real-time space operations · saai care
05 · Operating knowledge
This is not about producing more data. It is about answering why.
Stops, drop-offs and returns are read per zone. One axis alone will not reach the cause.
Pass-by · capture · entry · path · dwell · drop-off · gaze · pick-up · return · restocking · staff conversation · crowding · queue · stockout · anomaly
We cross the items against each other to see what happened and why. Scattered records become knowledge you can operate on.
Spatial analytics · saai insight
06 · The proposal
Analysis alone changes nothing. We put forward what to try.
Sales, inventory, weather and events from outside sit next to what the space told us. The agent looks for the answer before you ask.
A general AI gives a general answer. SAAI builds the structure and history of that one space, and answers from it. The person confirms.
Agentic AI for physical space · saai agent

Spatial · Anonymized · Agentic · Intelligence. That is SAAI.

Does the AI you use know your space?

Let us put them side by side

The difference

Numbers only record the outcome. SAAI knows what happened inside the space.

POStells you what sold, and how much.
General AIinfers the cause from those numbers.
SAAIreads what happened inside the store, and proposes the next action.

What each can do	SAAI	POS	Foot-traffic sensor	Video AI	General AI	Telco data	Transit card data	Manual count
Foot traffic, dwell time, and path	Can read	Cannot read	Cannot read	Limited	Cannot read	Limited	Cannot read	Limited
Entry and capture rate	Can read	Cannot read	Can read	Limited	Cannot read	Cannot read	Cannot read	Limited
Pick-up of an item	Can read	Cannot read	Cannot read	Cannot read	Cannot read	Cannot read	Cannot read	Limited
Payment amount and average ticket	Limited	Can read	Cannot read	Cannot read	Limited	Cannot read	Cannot read	Cannot read
Cause inference and ideas from the numbers	Can read	Cannot read	Cannot read	Cannot read	Can read	Cannot read	Cannot read	Cannot read
Next-action proposal in spatial context	Can read	Cannot read	Cannot read	Cannot read	Cannot read	Cannot read	Cannot read	Cannot read
Continuous analysis without storing raw footage	Can read	Cannot read	Limited	Cannot read	Cannot read	Can read	Can read	Cannot read
Can readLimitedCannot read

* Foot-traffic sensors count entries, not passers-by. Video AI has to retain the raw footage to analyze it. General AI can infer from POS numbers, but not which shelf produced them.

Telco data is a 50m grid estimated from cell signals. Transit cards count boardings, not who walked past your door. A manual count leaves only a few days of sample.

The method

We do not claim. We show the output.

Not a demo reel. Actual pipeline output.

Actual anonymization output

Anonymization SEAL

Identity is erased at capture. No privacy burden, so the cameras already on your ceiling are enough. After identity is removed, the shape of behavior in the space remains.

See anonymization

Actual MTMC output

Spatial Intelligence MTMC

We do not read one camera at a time. Separate views join into one store, seen whole. Connect movement across cameras into one continuous journey.

See spatial intelligence

Agentic AI

It reads the stacked signals and suggests what to check and which action to take next.

See Agentic AI

TRACK RECORD

Analysis continues across many spaces.

Analysis grows with incoming video

As of August 2026

60+
Customer sites
As of August 2026
14M hours of space read, and counting. More than 14,000,000 hours.
Video hours analyzed
Operating hours
150,000
People analyzed today
Estimated from operating volume
240,000
Actions analyzed today
Estimated from operating volume

Customer spaces

* Based on cumulative hours of operation across connected cameras. The figure on screen extrapolates at 1,073 hours per hour.

* Figures above are simulated, not measured. Assumptions: 19.8 sqm per camera · 66 sqm per convenience store · 300 daily visitors per store · 92s average dwell · 1 pickup per minute

NVIDIA Inception Partner · 103 patents

See it by industry

WHERE PEOPLE AND GOODS MOVE