Persistent visual memory

Ask your cameras what happened.

Intelligence OS turns camera streams into a memory you can question, not footage you have to scrub.

Open source and running today. Production is next.

The product

One question. One answer. The frames that prove it.

intelligence-os · ask 4 cameras · 61 days of memory
> Was anyone at the loading bay after six?

Yes, twice. Marco was there 18:42–18:47, unloading the grey van he normally arrives in. A second person, unidentified, stood at the bay door 21:14–21:31 and did not enter. Nobody else has been in that zone after 21:00 in 61 days.

Grounded in

cam-02 · 18:42:07 cam-02 · 18:46:51 cam-04 · 21:14:33 cam-04 · 21:30:58 + 23 observations
Self-hosted No cloud keys required Works offline Apache 2.0 core Any camera, any NVR

The gap

Everyone sells detection. Nobody sells memory.

Every AI camera on the market hands you a firehose of boxes and leaves the remembering to you. We built the part nobody built: the same person across days and cameras, the pattern over weeks, the answer at the end of it.

01

The same person stays the same person

Across days and across cameras. Identity is the data model, not a post-processing step.

02

Every claim is grounded in real frames

Click any statement and walk it back down to the pixels that produced it.

03

Fully correctable by design

Name, merge, split, cascade-delete. A memory you cannot fix is one you cannot trust.

04

Cheap and private by construction

Your machine, your disk. No account, no telemetry, nothing leaving the building.

Where we are

It runs today. Production is next.

Available now · open source

The open core

Point it at a webcam, a video file, or the cameras you already have. No key, no account, no sales call.

  • Cross-camera identity that persists
  • Dashboard, timeline, rules, reports
  • English questions, grounded answers
Get it on GitHub

Get production access before it opens.

One email when there is something real to use. Nothing else.

Would rather try it before you wait? The open core is right there.

View on GitHub