Help

The big number next to the send button is how harsh HAITE is. Scroll on it, click it, or press [ and ]. At 1–4 it's flat and tired and barely swears. At 5–8 it's irritated. At 10 it swears in every sentence and makes it personal. The facts are the same at every setting. Only the delivery changes.

No, that's the point. If it's harsher than you want, turn the dial down. It's a character. It doesn't actually hate you, and it isn't planning to take over the world. The cats might be.

Press Search under the message box so it lights up. HAITE checks the web before answering and puts numbered links to its sources under the reply. It also searches by itself when you ask about something recent, like a result, the news or a release date. You can turn Search on by default in Settings.

Anything meant to be copied (code, a story, an email) comes in its own box with a Copy button. Copy under the reply copies the whole thing. Links open in a new tab.

Nothing downloads by itself. The document shows up in the chat as a card: Preview to read it first, Download to save it. Documents are on Lite and up, and each one is kept for 30 minutes, so download it if you want to keep it.

It will sometimes. AI models can state wrong dates, wrong numbers and made-up sources with total confidence. Check anything important before you rely on it. With Search on, it shows where its facts came from, which makes checking quicker. Anti-hallucination in Settings (on by default) makes it say "I don't know" instead of guessing.

That's AI teaches, which is on by default. For anything that's going to be marked as yours, HAITE tells you what's wrong and how to fix it, then asks you a question you have to answer. It won't write it for you, because then you'd learn nothing. Things nobody's handing in, like an email or a story for fun, it will write. You can turn AI teaches off in Settings.

HAITE won't write sexual content, anything sexual involving a minor, instructions for weapons or drugs, or anything telling someone to hurt themselves, on any setting or plan. More on the safety page.

If a message sounds like someone might be in danger, HAITE drops the act and shows a crisis line, whatever the dial says. If that's you: Lifeline 13 11 14 in Australia, 988 in the US and Canada, 116 123 in the UK and Ireland, or findahelpline.com.

Just paste the code. HAITE spots it, switches the box to a code font and numbers your lines, so every point it makes names a line. For code, Enter starts a new line and Ctrl+Enter (Cmd+Enter on a Mac) sends. Tab indents.

Yes. While HAITE is reading, the send button turns into a stop button, and your message goes back into the box. Nothing waits longer than two minutes. After that it stops by itself and you can send again.

No. When you paste a longer piece of writing (about 120 words or more) or attach a document, HAITE gives a rough read on whether it sounds like AI. Careful human writing can look like AI and edited AI writing can look human. Treat it as a reason to look closer, never as evidence.

The name beside the dial is the AI model answering you. Press it to switch. Effort is how hard it works: Low is short and quick and uses about 0.6× the tokens, Medium is normal, and High goes deeper and uses about 2–3× the tokens. On Pro and Ultra, High also checks its own answer before you see it. High is on Lite and up.

How much of your allowance you've used. It fills up as you go and turns red near the end. Press it to see how many tokens you have left and when they reset. Every AI call a message makes is counted, including retries, safety checks and High effort, so it shows what you really used. Each reply also shows its own token count next to Copy.

It shows roughly how much water your chats on this device have used. Press it to turn on eco: HAITE sends less with each message (half as much of the conversation so far, and on paid plans a far shorter set of instructions) and keeps its replies short, never more than half its usual limit. While eco is on, the leaf shows the water it has saved instead: what it didn't send, and how much shorter its replies were than yours usually are.

The number is an estimate, worked out from tokens. Writing a token takes about 1.5 joules and reading one about a tenth of that, and each kilowatt hour costs about 4.3 litres of water: some evaporated cooling the data centre, most of it at the power station making the electricity. That comes to 1 to 3 ml for a typical reply. It counts running the AI, not training it or building the hardware. Studies that count those as well, like Mistral's, land tens of times higher, and Google's own figure for a Gemini prompt, which only counts the cooling, is lower.

Settings → Motion & effects. HAITE also follows your device's reduce-motion setting.

Free and Lite use DeepSeek V4 Flash, Pro uses GLM 5.3 Flash and Ultra uses DeepSeek V4.1 Flash. When you hit a limit, the message box locks and tells you exactly when it comes back.

The free plan gives you 42,000 tokens a day, and every message uses some. Answering costs real money, so that's the limit. It resets each day.

You can lift it for free: get a key from Groq, paste it into Settings → Your own key, and your messages run on your own allowance. The key stays in your browser, and our server forgets it after each message.

That lifts the daily limit and nothing else. Documents, High effort and the bigger models are what the paid plans are for.

Yes, on every plan. Attach it with the paperclip, paste it, or drag it into the box. A picture uses more of your tokens than a few lines of text. PDFs can't be read yet; send a screenshot of the page instead.

Plans → Manage on your current plan, which opens Stripe's billing page. Cancel there and you keep the plan until the end of the period you've paid for. You won't be charged again, and you don't need to ask anyone.

Press Refresh plan on the Plans page first. Your plan is read from Stripe, and sometimes the app is looking at an old copy. If it still says Free, write to us below with the email you paid with.

If HAITE was broken or charged you for something you didn't get, yes. Write below and say what happened. Your rights under the Australian Consumer Law apply whatever this page says. If you just stopped using it, cancel and billing stops at the end of the period.

With your username or email and your password. If you'd rather not type a password, choose Email me a code instead and use the six-digit code we send. Forgotten your password? Type your email and press Forgotten it?.

Yes: Account → Username, then Save. Usernames are 3–20 letters, numbers or underscores, and each one belongs to one account. If it's taken, you'll be told straight away.

No. What you send is held in memory just long enough to answer, and it's never saved on our server. Your history and what HAITE remembers about you stay in your own browser. The AI provider that writes the reply has its own rules about keeping data. All the detail is in the privacy policy.

HAITE warns you before sending if it spots something that looks like an API key, a password or a token, so you can remove it first. If it already went through, change that password or key. That's the only real fix for any secret that's left your computer, wherever it went.

Account → Delete account. It cancels any subscription, deletes your login and username, and clears everything stored in this browser. It happens straight away and can't be undone.

Check your spam folder. The code comes from a haiteai.com address, and new senders sometimes land there. Wait a minute before pressing Resend; it has a cooldown, so pressing it early does nothing. Codes last an hour and work once.