See it. Use it everywhere.

Point Vision at a picture and get something done: read a receipt, find a photo by words, or hear when your camera sees a delivery. Every answer says what it cost.

  • Free a read on our own engines
  • 8 tools, one place

Pick one thing to do

local first, AI when it helps

Reading, watching and the page check work without an account. Saving answers, your picture search and faces need an OpenVibe sign-in.

Your face choices · How faces are kept

Works across OpenVibe

8 items

Works across OpenVibe

The API
  • Pics and Mediaalt text for every image
  • Fooda receipt becomes your pantry
  • Dealsphotograph it, compare offers
  • Camalerts on your own cameras
  • Livescene changes and moments
  • Designthe page check
  • Searchsearch by picture
  • Benchwhich model sees best, per dollar

What Vision does

A receipt, read for nothing

The local reader (Tesseract, on this host) reads the words a receipt actually needs: merchant, date, lines, total. No image leaves the machine and no model is called. Only when the local pass finds no usable text does a signed-in person's AI fallback run, and the answer says so.

Read a picture
Example · local OCROPENVIBE MARKETa dateOat milk 3.40Bread 2.10TOTAL 5.50example values · not a measured result

The same image, side by side

A compare sends one image to the local reader and to OpenVibe.AI. AI's router is the authority: the provider and model shown are the ones that answered, never the hint. Cost comes from AI's own run, latency is measured here.

Run a compare
Example: one image, three answers
engineanswercosttime
local OCR“TOTAL 5.50”0.000038 ms
pool:fasta receipt from a market0.00041.1 s
pool:bestthe total is 5.500.00122.4 s

The UI check catches the overflow

Drop in HTML, a page URL or a screenshot. The static pass reads the boxes and the colours: a container wider than the viewport, nowrap text clipped by its box, body text under 4.5:1, positioned boxes on top of each other, a z-index with no effect. Arena calls the same API after it generates a page.

Check a page
Example
this card is 1400px widea headline that gets cut off at the edge
  • overflow1400px in a 390px view
  • cut offnowrap inside 120px
  • contrast2.4:1, needs 4.5:1

Faces, only with consent

Each face feature is its own switch with its own consent: verify yourself, find yourself in your own pictures, tag co-streamers who enrolled, or group the people in your own library. Matching runs on this host; templates are encrypted and erased the moment you switch a feature off. Detecting, counting and blurring identify nobody. Vision never identifies a stranger.

Your choices

Questions

3 questions

Questions

Does Vision recognise faces?

Only for a person who separately opts into a specific purpose: self verification, self search, cast tagging or private library grouping. Detect and blur never identify anyone. AI may describe what it sees, people included, but it never names a stranger or uses a face as identity.

Which model answers?

Every answer names the model and provider that actually answered and the reason it was chosen. Local OCR is first for a read question; otherwise OpenVibe.AI routes the run and its provenance is what Vision reports.

Are my images stored?

General answers keep no image bytes. An opt-in face library stores encrypted uploads and vectors privately; turning its purposes off erases derived data. Account export and deletion include those rows.