Privacy Policy

What we collect

An account requires an email address. You sign in with a one-time code we email you. Creating a listing means giving us the vehicle details you enter and the photographs you upload. A guided capture session in our phone app is also recorded, video and audio; recordings are available to the admins on your account, and to our own staff and the systems that run the service, and are kept for up to 90 days, then deleted, as described below. We do not collect payment card details: there is no billing system in the product today, and if paid plans launch, this policy will be updated before any card is charged.

Your photographs

Uploaded photos are processed to separate the vehicle from its surroundings and to produce the finished versions shown on your listing. Both the originals and the results are stored so your listing keeps working. We do not sell them. The companies that handle them for us are named under Service providers. Deleting a vehicle deletes its images from our storage. Working copies made while we process them, on computers we operate, are not yet removed automatically; we remove them on request.

The capture app on your phone

Our iOS app guides someone around a vehicle and photographs it. It asks for three permissions, set out below, and it records each guided capture session.

Camera. It takes the photographs, and it also sends a low-resolution video stream to the guide while you walk so the guide can tell you where to stand and what it can see. Individual frames from that stream are sent to a third-party model, described below, to produce those spoken directions. Some of those frames are also stored privately with the vehicle, like your photographs, and deleted when the vehicle is deleted. The same video is part of the session recording.

Microphone. You can talk to the guide and it answers out loud. The guide hears your voice and sees the live camera: your voice is streamed live, with the camera video, to a third-party realtime model for the length of the session, and both are part of the session recording.

The session recording. A guided capture session is recorded: the camera video while the phone is sending it, and the audio in the session, which includes the person holding the phone, anyone else the microphone picks up, and the guide. Recordings are stored privately and are available to the admins on your account, and to our own staff and the systems that run the service. They are kept for up to 90 days, then deleted. Deleting the vehicle deletes its recordings from our storage sooner.

Typed questions. If the live voice is unavailable, you can type a question instead. Some questions are answered from the app’s own fixed replies. When none fits and the phone has attached a still from the camera, the text you type and that still are sent to a third-party model, OpenAI’s, to produce the answer. The questions you type, and their answers, are kept with the vehicle’s history and deleted when the vehicle is deleted.

Photo library. Only to save a copy of the photos the app took, if you ask it to. It never reads your library.

The LiDAR scanner, on phones that have one, measures the ground and the distance to the vehicle so the app can place the walking ring and check its own coverage. We are specific about where that goes because the natural assumption is wrong: the depth image is uploaded and stored alongside the photograph, not merely used on the phone and discarded. It is a distance measurement of the ground and the vehicle in front of the camera, and it is treated like your photographs: not sold, and deleted when the vehicle is deleted.

Not location. The app contains no location code at all. It does not ask for, receive or store where you or the vehicle are.

Training, and your choice

We build our own vehicle imaging models rather than renting someone else’s. Those models improve when they learn from real dealership photography, which raises an obvious question about who decides.

Processing is not training. We process your photographs, and frames from the live camera during a capture session, to make your images: to find the vehicle, its windows and its inside parts, and to build the finished photos and 360 view. That work uses our software, and the companies named below, but it does not teach any model anything. Checking finished results, and measuring how well our software handled them, is also not training and does not change a model.

You decide about training. We do not use your photographs or your session recordings to train our models unless you turn that on. Training is off by default and stays off until you choose otherwise. Business customers can grant it in writing as part of an agreement, which is where the terms, the scope and any compensation belong. You can withdraw permission at any time.

One thing we will not pretend about: withdrawing stops any future training and removes your images and recordings from our training sets, but it cannot pull them back out of a model that has already been trained. Retraining without them happens on our next scheduled cycle. Anyone who tells you otherwise is describing something machine learning does not currently do.

Where your images are processed

Your photographs are kept in private storage, not in a public folder. The people on your account can reach them, and so can our own staff and the systems that run the service. While we process them, working copies also sit on computers we operate. To turn them into finished listing photos and 360 views, we run our own image software. Most of it runs on rented graphics computers at RunPod. One part, the check the capture app makes on each photo as it is taken, runs on a computer we operate. A few steps also use services run by other companies. Each of those is named below, with what it receives and why.

Number plates, people and other cars near the vehicle can appear in the photographs and live camera frames we process, including those that go to OpenAI and Meta. Where you can, move people out of frame.

The public demo pages on this site work differently. A photograph you upload there is sent to a computer we operate, which separates the vehicle from its background and returns the result. The photograph is not sent to OpenAI or any other outside service for processing, and our software on that computer keeps no copy of it. What you type into the demo’s research tools goes to Firecrawl, and a VIN or a vehicle’s year, make and model goes to the US vehicle database, as described under Service providers. Demo uploads are never used for training.

Service providers

These companies process information for us. Each receives only what it needs for its part. What each one does with that information is governed by its own terms; we send them only what is listed here.

• Supabase — our database and private file storage: accounts, vehicle details, your photographs, the depth measurements, session recordings and the finished results. It also sends your one-time sign-in codes by email, using Resend to deliver them.
• Vercel — hosts the website and the programs behind it, so everything you upload or view passes through it. Its analytics tools count page views and measure how fast pages load, and a bot check protects the public request form.
• RunPod — rented graphics computers on which our own software separates the vehicle from its surroundings, adds the studio and renders the 360 view and listing photos. It receives your photographs and working images made from them through short-lived private links, and frames from the live camera during a capture session through our own servers; the results are written back to our storage.
• A computer we operate — the check that the capture app makes on each photo as it is taken runs on a computer we operate. It receives that photograph, and it may also receive frames from the live camera during a capture session. It also receives a photograph uploaded in the public demo pages.
• OpenAI — reads your photographs to note things that affect the finished images, such as dust, reflections and what surrounds the car, and reads a window sticker if you photograph one. For each photograph it receives the photograph and the vehicle’s VIN, year, make, model and trim. We send these requests with the setting that asks OpenAI not to store the response. OpenAI also runs the live voice guide, which hears your microphone and sees the live camera during a capture session, including the frames the phone sends while the camera is open. It answers the questions you type in the capture app when no fixed reply fits, receiving the text and, if the phone attached one, a still from the camera, sent directly to OpenAI. While the capture app checks that a window or an inside view is lined up, it also sends OpenAI a still from the live view. And it writes your listing and social text from the vehicle’s facts and the details you enter, without photographs.
• Meta — outlines the windows and the inside parts of the car in your inside and window photographs. It receives each photograph, or a crop of it, and a short word such as “window”. While the capture app checks that a window is lined up, it also sends Meta stills from the live view.
• LiveKit — the live audio and video connection between the phone and that guide. It carries the media and makes the session recording, which it uploads to our private storage. It also hosts the guide software. It is not what generates the guidance.
• Firecrawl — a web search service we use to look up your vehicle, such as its window sticker, equipment and recalls. It receives the VIN and the vehicle’s year, make, model and trim. For market comparables it also receives the ZIP code and search radius you give, and if you paste in a listing it receives that web address and returns the page. In the public demo pages it receives the question or search you type. It does not receive your photographs.
• US vehicle records and manufacturer sticker services — to decode a VIN and find recalls we ask the US government vehicle database (NHTSA), and for some makes we ask the manufacturer’s public window-sticker service. They receive the VIN or, in the public demo pages, the vehicle’s year, make and model.
• Resend — sends our emails: sign-in codes, team invitations, notices of enquiries about your listings and demo requests. It receives the recipient’s address and the message.
• Upstash — counts requests so nobody can flood our forms and tools. It stores counters tied to an account, a session or a scrambled form of an address or email, never photographs. It may also keep, for a few minutes, a copy of a vehicle list (details and image addresses, never the image files) so that lists load quickly.
• Sentry — error monitoring. When something breaks on a page, it receives the visitor’s IP address, browser details, the address of the page and the addresses of requests the page makes, with access tokens removed. For some visits it also receives a replay of the page, in which every image and all text are blanked out. This runs in your browser only; our own servers send it nothing.

People on the account can open the live view of a session in progress and watch the camera as it happens. The live view has no sound. Recordings of a session, which do include its audio, are available to the admins on your account, and to our own staff and the systems that run the service. They are kept for up to 90 days, then deleted. Everyone in a capture session should assume the people on the account can watch it.

Public listings

Publishing a vehicle makes its details, photographs and asking price publicly visible on the marketplace and on its own listing page, which search engines may index. Your email address is not shown; buyer enquiries reach you through the site.

Your choices

You can edit or delete any listing from your dashboard, and you can ask us to delete your account and everything attached to it. Write to privacy@autolensai.com and we will action it.

Changes

If this policy changes in a way that affects how we handle your information, we will update the date above and notify account holders by email.