AI Use and Your Camera

Ask — the mission assistant

You type a question and we answer it using the launch data we already hold. Your question is sent to our servers, combined with the relevant mission, rocket, pad or agency records, and passed to a language model.

The answer is grounded in that data on purpose: the model is given the records and asked to answer from them, rather than from whatever it remembers about spaceflight. It is still a language model, and the section below on being wrong applies.

Your conversation is stored against your account so it survives closing the app. You can delete a conversation, and deleting your account deletes all of them.

What is sent: your question, the conversation so far, and the launch data we selected as context. Your app identifier and an internal numeric user id.

Explain — one-shot analysis

Tapping Explain on a launch, rocket, spacecraft or pad sends that entity's public record and asks for a plain-language summary. It carries no conversation history and nothing about you beyond an internal numeric user id.

Analyze Scene — camera settings

You point the camera at the sky and tap Analyze Scene. The app captures one frame and sends it to our servers along with numbers describing your current setup: shutter speed, ISO, exposure compensation, white balance, which lens you are on, what your device is capable of, and measurements of the scene's brightness and clipping. It also sends the latitude and longitude you are observing from, so the server can work out where the Sun and Moon are and how dark the sky is — that geometry is what makes the exposure suggestion any good.

Our server passes that to a vision model, which returns suggested camera settings, a short explanation of why, a confidence score, and some tips. We clamp every suggestion to what your specific device can actually do before showing it to you, so it cannot suggest a shutter speed your camera does not have.

You then decide whether to apply it. Nothing changes on your camera unless you say so.

### What is sent, precisely

• One still frame from the camera preview.
• Your current camera settings and the device's capability limits.
• Scene measurements: brightness, highlight clipping ratio, trail occupancy.
• The latitude and longitude you are observing from, and the local time, so the Sun and Moon can be placed. Only when you have granted location access.
• Your app identifier and a request id.

### What is NOT sent

• Your photo library. Not any of it.
• Your recorded video.
• Your location in the background, or at any time you did not tap Analyze Scene. The position is read once, for the frame you asked to analyse, and never continuously.
• Your name or email address. The request carries an internal numeric user id.
• Any other frame. One frame, when you tap the button. There is no background capture and no continuous stream.

### Whether the frame is kept

By default the frame is not retained. It is held in memory long enough to be analysed and then discarded. We store the *result* of the analysis — the suggested settings and the reasoning — not the image.

Frame storage exists as an option we can switch on for diagnostics, and when it is off, which is the default, no frame is written to disk anywhere in our system. Each analysis record carries a flag saying whether its frame was stored, so this is auditable rather than a promise.

If we ever change the default, that is a change to this document, which means the app will ask you to accept it before it takes effect.

Read Aloud — speech

When you ask to hear an answer, the text of that answer is sent to our speech provider and the audio is streamed back to you. The audio is not stored — not by us, and not as a file anywhere in our system. Your voice is never involved: this feature speaks, it does not listen.

Your data is not used to train models

We do not train models on your data. We do not fine-tune on it. Our AI provider processes each request under a zero-retention arrangement and does not use API inputs to train their models.

We are aware this is the single most common worry about AI in an app that can see your camera. The answer is no.

The AI can be wrong

Every feature above is a model making an educated guess. None of them is a measurement or an authority.

The assistant will sometimes state something confidently and incorrectly — a date, an orbit, a payload. Do not use it to decide when or where to be for a launch. The launch times, windows and pad locations shown in the rest of the app come from spaceflight data sources and are what you should plan around; the assistant is for understanding them, not for scheduling your drive.

Analyze Scene has the same limits any single-frame guess has. It will occasionally recommend an exposure that blows out your highlights or misreads an unusual sky. You are the photographer. Do not rely on it for a shot you cannot retake — and a launch is the definition of a shot you cannot retake.

Metering, and why it exists

Each request costs us money, so the AI features are metered:

• Free — a one-time starter allowance for the lifetime of the account, not monthly: 25 assistant credits, 5 scene analyses, and 300 seconds of speech.
• Pro (any subscription, or the lifetime unlock) — a monthly allowance that refreshes each period and does not roll over.
• Credit packs — purchased credits never expire. We spend your monthly allowance first and your purchased credits only after it runs out, so you are never charged for something your subscription already covered.

When you run out, the app tells you what you have used and what your options are. It does not silently stop working.

Turning it off

Every AI feature here runs only when you tap it. There is nothing to disable — not using one is the same as it being off. Nothing is sent in the background, no frame is captured unless you ask for one, and the assistant does not read your conversations to do anything other than answer your next question.

Changes

We will publish a new version of this document if any of the above changes, and the app will ask you to accept it. The specific commitments in this document — one frame only, on your explicit action, not retained by default, audio not stored, nothing used for training — are the ones we consider material, and we will not weaken them quietly.