1. Features
  2. Private AI
New in 0.2

Private AI that never leaves your computer

Private AI reads your recordings on your own computer and tells you who was there and what happened, in plain sentences. It's off until you turn it on, it only looks at the cameras you choose, and it sends nothing anywhere.

  • 0Pictures or results uploaded
  • 15%Of one CPU core by default
  • 10sBetween pictures it describes
  • 5Face samples kept per person, at most
The Activities page in the Camlane desktop app with a written summary of the day per camera and an activity log.
Activities: a summary of the day per camera and per person, and the log behind it.

Off until you say so

Private AI has several switches, each off by default:

  1. Private AI itself, in Settings → Private AI (fully offline).
  2. Each camera, one by one.
  3. Recognise people, and again per camera.
  4. Suggest what people are doing.

While it runs, the sidebar says “Private AI on”, and adds “people” while it recognises faces. It works only on recordings already on the computer’s disk, never on the live stream.

How it works

The first time, Private AI downloads its models, each only after you accept its licence, and checks every file before use. On a computer without internet you can copy the files in instead. From then on it works fully offline, at the lowest priority and within a CPU budget (15% of one core by default), describing a picture from your recordings every 10 seconds so you can find people and moments later.

People library

  • Like a photo library. Faces are grouped by person. Name someone once and their earlier and later appearances get the name. Unnamed groups appear as “Unknown 1”, “Unknown 2” and merge when they turn out to be one person.
  • Consent first. Camlane lets you name a person only after you confirm they know the camera recognises them and agree, and it records when you confirmed.
  • Little is stored. At most five different face samples per person, as templates (numbers, not photos). A picture of each person is kept only if you turn on “Keep a picture of each person”.
  • When and where. For each person: a month heatmap, their visits for a day (“came 09:10–10:40”), time on camera, and Play to jump into the recording.
  • Delete person removes their templates, pictures, name, sightings, tags and notes.

Activities, highlights and day summaries

  • What people were doing. Short stretches of a person’s appearances are compared with activity words (“cooking”, “on a call”, “going out”) and with what you tagged before. You can add your own words.
  • Tag this? When it can’t tell, it asks on the Activities page, with a picture and a short clip, at most once per person and camera every 15 minutes. Your answer also tags look-alike moments, so suggestions improve.
  • Highlights. “Save highlight…” in the player keeps a moment with a picture and your note, and keeps its recording until you delete the highlight.
  • Day summaries. Plain sentences such as “Sam was here from 09:10 to 10:10, cleaning and cooking.” They’re filled in from what is stored each time you look, not written by a language model or an online service.

Your data, your rules

Everything that describes a person is encrypted with the recordings key. It follows the recordings: when a recording is deleted, what Private AI learned from it goes too. Delete all AI data removes everything at once.

Faces count as biometric data in many places, including under the GDPR, Illinois’ BIPA and India’s DPDP Act. Tell people that a camera records and that recordings are analysed, and add someone to the people library only with their agreement. The privacy policy has the details.

Guides