A flag is a lead. A person decides.
Knolt flags. Your team decides. Your footage stays yours.
- ID
- #19
- Description
- dark jacket, backpack
- Journey
- hall → corridor → vault
- Time
- 21:14:32
- Flag
- To review
Never
NameAgeGenderEthnicity
Five rules we do not bend.
- 01
A person decides
Knolt never acts on its own.
- 02
Your footage stays yours
In your cloud or the branch. Never sold.
- 03
Faces are leads, not names
On your box. Audited. Can be switched off.
- 04
Only what is needed
Never age, gender or ethnicity.
- 05
Measured, not promised
Tested on unseen footage before every release.
Every action, on the record.
Real Knolt output

What Knolt does, and what it never does.
Knolt
- Numbers each person in view
- Describes clothing, for search
- Links cameras; a supervisor confirms
- Compares faces where you allow it
- Audits every face search
Knolt never
- Acts without a person deciding
- Identifies anyone by their face
- Guesses age, gender or ethnicity
- Moves a bank’s video out
- Sells your footage
Never a verdict. Never a name. Never sold.
The numbers, and where they come from.
- of robberies caught
- 0%
of robberies caught
On footage Knolt had never seen.
- false alerts
- −0%
false alerts
After tuning to the branch.
- same answer on every replay
- 0%
same answer on every replay
Same footage, same call.
95% is whole robbery incidents caught on unseen footage. The 83% on the comparison page is Security-Bench 1.3: a correct call on real security-camera frames and staged scenes, the test every model got.
See the comparison →