Getting started
Five steps, about ten minutes. Four of them you do once.
What you need
- the calculator
- a 2.4 GHz WiFi network — see the warning in step 3
- an OpenAI account with a little credit on it
1. Sign in
Type your email on the sign-in page. A link arrives by mail; clicking it signs you in.
There are no passwords anywhere in this system. Nothing to forget, nothing to reuse, nothing to steal from us.
2. Add your OpenAI key
Go to API key and paste a key from platform.openai.com. It is checked against OpenAI
before it is saved, so a typo is caught while you are looking at the field rather than twenty
seconds into a shot.
Requests are billed to your own OpenAI account. A few things follow from that, and they are worth knowing:
- The key never reaches the calculator. It stays on the server, encrypted.
- The site cannot read it back, and that is enforced by AWS rather than by our code: the identity this site runs under is allowed to encrypt and is not allowed to decrypt. Only the request handler can. So there is no "show key" button, and there will not be one — only replace and delete.
- Nothing is stored in plain text. A dump of our database without the key from AWS is a bag of noise.
3. Put the calculator on your WiFi
Press MODE, choose 3 WIFI SETUP, and type the network name and password on the
on-screen keyboard. MODE cycles between the fields; the dedicated key sends them.
Two things that catch people:
The radio only sees 2.4 GHz. That is the hardware, not a setting. If your router
broadcasts two names — something like Home and Home_5G — you need the first one. If it
broadcasts one name on both bands, the calculator will find it anyway.
Typing a password one letter at a time is tedious, and we know it. It is the worst part of the setup and it is on the list to fix.
4. Pair the calculator
Press MODE, choose 2 AI MODE. The screen shows PAIR CODE and five characters.
Type those five characters on the Devices page here. Within a few seconds the calculator picks up its credentials and is ready — you do not need to press anything on it.
The code is good for 15 minutes. If it expires, leave AI mode and come back; a fresh one appears.
5. Take a shot
In AI mode, aim at some text and press the shoot key.
The calculator says hold still for about four seconds — that is the camera finding focus,
and moving during it produces a blurred frame. Then got it, can move: the picture is
taken and you can lower the calculator while the answer is on its way. About twenty seconds
later the answer is on the screen; ▲ and ▼ scroll it.
The whole request now appears under History — the photo, the answer, the tokens and what it cost.
Choosing a model
MODE → 4 GPT switches between two models. The difference is real:
| model | what it is | a typical shot |
|---|---|---|
gpt-5.6-luna | fast, cheap, the default | ~$0.0005 |
gpt-5.6-terra | better at hard questions | ~$0.005 |
Ten times the price for the same photo. Measured on the same frame: 2418 tokens cost $0.00049 on luna and $0.00489 on terra. For reading text off a page, luna is usually enough; terra earns its money on maths and on questions that need reasoning rather than reading.
When something goes wrong
The calculator shows the reason in words rather than a code, because the wording comes from the server and can be improved without touching the firmware.
| on screen | what happened |
|---|---|
NO API KEY - SET ON SITE | no key on your account yet — step 2 |
API KEY REJECTED | OpenAI refused the key; it may have been revoked |
DEVICE NOT PAIRED | pairing was revoked, or the calculator was reset — step 4 |
PAIR CODE … | not paired yet — step 4 |
AI: too dark, add light | the camera could not get a frame at all |
AI: no answer | the module went quiet for a minute; press AC and try again |
If the calculator stops responding entirely, ON switches it off and on again.
What survives that, and a flat battery too: your WiFi credentials, the prompt you
typed and the model you chose. They are kept in memory that holds its contents without
power, so you never set those up twice. Treat anything else — the expression on the screen,
the M memory, the last answer — as gone once the calculator switches off.