Caufox Documentation

Overview

Caufox is an AI chat application that runs entirely in your browser. You can ask questions about any topic, request complete code for websites and scripts, let the AI read and summarize links, and save every result as a file that you can preview, copy, download, or share with others. Everything you type is streamed back to you in real time, so the experience feels instant and natural.

The application was created and is maintained by Caufox Indonesia. It combines three AI models, a transparent token quota system, local QRIS payments, and full file management into one simple chat interface. Whether you are a student writing an essay, a developer prototyping a landing page, or a business owner asking for marketing copy, Caufox is designed to be useful without requiring any technical setup.

• 3 AI models — Luona (default), Tide (deep analysis), Halo (reasoning).
• Real-time streaming — answers flow in live and can be stopped at any moment.
• Automatic files — long answers become file chips with a one-click RUN preview.
• Web tool — the AI opens links you paste and uses their content as context.
• Flexible plans — Free, Pro, and Max with a clear 5-hour token quota.
• QRIS payments — scan and pay with any Indonesian bank app or e-wallet.

All features are available on every plan. Plans only change how many tokens you can use and how long each answer can be. There are no hidden fees, no credit cards required, and no complicated configuration.

Quickstart

Getting started with Caufox takes less than a minute. You do not need to install anything or create a complex profile — an email address is enough. Follow these five steps and you will be chatting with AI immediately.

1. Open caufox.com in any modern browser on your phone or computer.
2. Click Sign in in the top navigation, then enter your email address.
3. Check your inbox for a 6-digit OTP or click the magic link we send you. The link works once and expires after one hour.
4. Once logged in, you land on the chat screen. Type your first question and press Enter.
5. For better answers, raise the effort level to medium or max before sending.

If you want to see the file feature in action, try asking: "Create an HTML landing page for an online apple store, responsive, with a WhatsApp button." The AI will reply with a short explanation and a downloadable file chip.

Your chats are saved automatically in the sidebar. You can start a new conversation at any time with the New chat button, and continue an old one later by clicking its title. Nothing is lost when you close the tab or switch devices, as long as you log in with the same account.

Create Account

Creating an account on Caufox is intentionally simple. You only need a valid email address — there is no mandatory username, no phone number, and no payment information at registration. The whole process is designed to take under a minute so you can start using the service immediately.

• Email verification — you receive a 6-digit OTP code or a single-use magic link. The magic link is valid for one hour.
• Profile — after signing in you can set a nickname from the settings page; it is used in the chat interface and sidebar.
• Data stored — your account holds your email, nickname, display preferences (dark/light theme), saved memory, and chat history.
• Security — your session is protected by a token that you can revoke from Settings → Sessions at any time.

You can use the free plan forever with no time limit. If you later decide to upgrade to Pro or Max, you pay only via QRIS — no international card required. The same email that you registered with is used to link your plan, your chat history, and your files, so everything stays in one place.

If you ever lose access to your account, request a new magic link from the login page. A fresh one-time link will be sent to your email. For security reasons, links cannot be reused, and we never send passwords over chat.

Login & Magic Link

Caufox offers two ways to sign in: a password and a magic link. The authentication pages are written in full English for consistency. Both methods are secure, and you can switch between them freely.

• Magic link — enter your email and we send a one-time link. Opening it logs you in instantly without typing any password. The link expires after one hour and can only be used once.
• OTP code — a 6-digit code is sent to your email. Enter it on the page to complete the login. Please request the code only once — spamming the resend button delays delivery.
• Password — for returning users who prefer the classic flow, the password option works like any standard account.
• Sessions — every active session is listed in Settings → Sessions. You can see the device, IP, and time, and revoke any session remotely.

When you log out, your local data (conversation cache and token) is cleared from the browser, but your account data remains safely on the server. Logging back in restores your history automatically.

Tip: if you use the magic link from a phone and the browser opens, the app logs you in automatically and returns you to the chat. If the link shows an "invalid or expired" message, simply request a new one.

First Chat

Your first message sets the tone for the whole conversation. Caufox sends the last ten messages as context, so the more specific you are, the better the answer. Here is a practical example that shows off both the explanation and the file features:

"Create an HTML landing page for an online apple store called
Apel Fresh, responsive, with a WhatsApp button"

The AI will reply with a short explanation of what it built, followed by an HTML file chip. Click the chip to open a live preview (RUN), use the ⋮ menu to copy the code, save it as HTML or PDF, open it in a browser, or copy a share link. The file is also uploaded automatically to private object storage, so you can retrieve it later.

For everyday questions, simply type naturally. You can ask follow-ups, and the AI keeps the thread. If you want to change the direction of the conversation, start a new chat — the history stays available in the sidebar.

Remember the effort setting: low gives the fastest answers with no thinking step, while max produces deeper reasoning for complex problems. You can change it at any time, and it is remembered per device.

Chat Basics

The chat interface is designed to be as familiar as any messaging app. Messages appear as bubbles, your messages on one side and the AI's on the other. The input bar sits at the bottom and supports multi-line text, file attachments, and voice input.

• Send — press Enter to send; Shift+Enter creates a new line.
• Attachments — click the 📎 button to attach a text file (txt, md, json, code). The file content is inserted into your prompt, up to a sensible size limit.
• Context — the last ten messages are included as context for the AI, so follow-ups make sense.
• Feedback — every answer can be liked or disliked, and retried with the retry button.
• Message actions — long-press one of your messages to copy it or edit it.
• Scroll — the message list auto-scrolls to the newest message while the AI streams.

The input supports everything you would expect: emojis, code, long paragraphs, and links. There is no message-length limit in the input itself — long prompts are handled gracefully.

If the connection drops, Caufox retries automatically up to three times before showing an error, so a weak signal does not ruin your conversation.

Models (Luona / Tide / Halo)

Caufox gives you a choice of three AI models. Each one is tuned for different kinds of work, and you can switch at any time from the picker below the input bar. Your selection is remembered for the next time you visit.

Luona is the recommended default for most conversations: it responds quickly and handles a wide range of topics well. Tide shines when you need thorough, well-structured long-form content — research summaries, essays, or detailed guides. Halo is the model to choose when a problem requires careful reasoning, such as debugging logic, math, or decision analysis.

All models share the same features: streaming, file generation, web tool, and memory. Switching models mid-conversation is allowed; the AI will pick up the context from the message history.

Tip: for code generation, most users prefer Luona or Tide. For multi-step reasoning, try Halo with effort set to max.

Choosing a model is not permanent — every message can use a different one, and the context carries over seamlessly. This flexibility lets you match the tool to the task instead of adapting the task to a single model. The picker also shows which model is active so you always know who is answering.

For teams or repeated workflows, a consistent choice (Luona for general, Tide for reports, Halo for logic) makes results predictable.

Effort & Thinking

The effort control decides how much reasoning the AI performs before answering. It is a simple way to balance speed against depth, and it applies to every model.

• Low — no thinking step. The answer streams immediately. This is the fastest option and perfect for quick questions, greetings, or simple facts.
• Medium — the AI runs a short thinking phase with a visible Thinking… bar. Good for everyday questions that benefit from a little structure.
• Max — full reasoning with dynamic sub-steps. Best for complex code, analysis, and anything that needs careful thought.

During thinking, a standalone bar appears with animated dots and, on medium/max, sub-step titles chosen by the model itself. The bar is clickable, and thinking content is not saved to your chat history — only the final answer is stored. This keeps your history clean and readable.

The thinking phase is intentionally kept short so you are never left waiting: it is capped at a few rounds and a few seconds. After thinking finishes, the answer streams in live.

Your effort choice is stored per device and stays active until you change it.

Streaming & Stop

Every answer on Caufox is streamed: tokens appear on screen as the AI generates them, instead of waiting for the full response. This makes long answers feel instant and lets you read the beginning while the rest is still being written.

While the AI is answering, the send button turns into a Stop button. Clicking it halts generation immediately. Whatever was produced up to that point stays on screen, and the partial answer is saved to the conversation.

Closing the tab or navigating away does not cancel the process on the server. The conversation continues and is saved, so when you come back, your messages and the completed answer are there.

Streaming also applies to the thinking phase: sub-steps appear as they are discovered. If the connection is interrupted, Caufox retries up to three times with a short delay, then shows a clear error message instead of leaving you hanging.

Streaming changes how you interact: you can read the beginning of a long answer while the rest is still being generated, spot errors early, and stop generation the moment you have what you need. The partial content is never lost — it remains in the conversation and can be copied or edited.

On slow connections, streaming still delivers the first words quickly, making Caufox feel responsive even under poor network conditions.

Files & Code

One of Caufox's most useful features is automatic file handling. Instead of pasting huge walls of code into the chat, long outputs are turned into neat file chips that behave like real files.

• Thresholds — text answers longer than 400 characters, or code longer than 300 characters, automatically become a file chip.
• RUN preview — clicking a chip opens a live preview. HTML is rendered in an iframe; PHP is simulated (tags like <?= ?> and echo become text, the rest renders as HTML).
• Menu ⋮ — Preview, Code, Copy, Save as HTML, Save as PDF, Open in Browser, and Copy Link.
• Storage — files are uploaded to private S3 object storage automatically (☁ badge), so they are safe and shareable.
• Short code — smaller code blocks get a COPY / VIEW / EDIT toolbar above them instead of a chip.
• Icons — HTML 🌐, PHP 🐘, JS ⚙️, everything else 📄.

File names are derived from the content (for example landing-page.html) when the AI provides one, or generated generically otherwise. The extension is detected automatically, so previews always use the right renderer.

You can also share files: the Copy Link action creates a private link that works until the file is removed. This is useful for sending a generated page to a colleague or client.

Web Tool

The web tool lets Caufox read web pages for you. When you include a link in your message, the AI opens that page during the thinking phase and uses its cleaned-up content as context for the answer.

"Open https://example.com and summarize it"

The fetched HTML is stripped of scripts, styles, and navigation clutter before being passed to the model, so the AI works with the meaningful content of the page. This is useful for summarizing articles, checking product pages, comparing information, or extracting key points from documentation.

There is no need to copy-paste the text yourself — just paste the URL. The web tool works with any public page. If the page is unreachable, the AI will tell you and answer without it.

Tip: combine the web tool with the file feature: ask the AI to read a page and then write a summary file, or generate a landing page inspired by an existing website.

The web tool is safe by design: only the page's readable content is extracted, scripts and trackers are stripped, and the fetched text is used solely as context for your answer. You keep full control — the AI never follows links inside links unless you ask it to.

Practical uses: summarizing news, comparing product specs, extracting contact info from a page, or checking documentation for a quick answer.

Clarification

Sometimes a question is genuinely ambiguous. Instead of guessing, Caufox asks you directly with a clarification popup that shows a question and a few answer options.

Example: you type "apel" — the AI asks "Do you mean the fruit or the brand?" with options like [Fruit] [Brand] [Other]. You can tap an option or type your own answer. The popup is styled to match the app theme and works on both desktop and mobile.

This keeps conversations on track: the AI never wastes a whole answer on the wrong interpretation. The clarification flow is used sparingly — only when the ambiguity is real and material to the answer.

If you would rather not answer, close the popup and rephrase your question with more detail.

Clarification is not a limitation — it is a quality feature. By resolving ambiguity before answering, Caufox avoids the common problem of a confident but wrong answer. The popup appears only when the meaning genuinely matters, so you will not be interrupted for trivial details.

You can also preempt the popup by adding context yourself: instead of "apple", write "apple fruit vs Apple brand". The AI picks up on the detail and skips the question entirely. When the popup does appear, the options are generated to cover the most likely interpretations, and the "Other" option lets you type anything.

Memory

Caufox can remember important facts about you across conversations. When the AI notices something worth keeping — like your name, your project, or a preference — it saves it as a memory entry tied to your account.

• Automatic — memories are extracted from your conversations in the background.
• Injected — on your next chat, relevant memories are added to the prompt automatically, so the AI remembers who you are.
• Manageable — open Settings → Privacy to see every memory, edit its value, or delete it entirely.
• Instructions — you can also set custom instructions that the AI follows, as long as they do not conflict with system rules.

Memory is private to your account. No other user can see it, and it is never shared. You are in full control: deleting a memory removes it from future prompts immediately.

Tip: tell the AI things like "I prefer concise answers" or "I am building a food delivery app" — it will remember and adapt.

Memory grows with your usage: the more you chat, the better the AI understands your context. Extraction happens in the background on every conversation, so nothing requires manual effort. You stay in control with the privacy panel — inspect every stored fact, correct it, or clear it all with one action.

Memory is separate from chat history: deleting a chat does not delete memory entries, and vice versa.

Voice Input

Typing is not always convenient. Caufox includes voice input so you can dictate your messages, especially useful on mobile.

Click the 🎙 icon next to the input bar. On supported browsers, Caufox uses the built-in speech recognition with Indonesian language support, transcribing your speech as you talk. When you stop, the text appears in the input box and you can edit or send it.

If your browser does not support speech recognition, Caufox automatically falls back to recording your audio and transcribing it on the server. You will see a "Transcribing..." placeholder while the audio is processed, and the result appears in the input.

Voice input respects your privacy: audio is used only to produce the transcription and is not kept afterward.

Voice input is especially handy in three situations: on the go, when your hands are busy, and when you are dictating longer thoughts that would be slow to type. The transcribed text lands in the input box, where you can still edit it before sending — nothing is sent until you press Enter.

The server-side fallback uses audio recording and transcription, which works on more browsers than the built-in speech recognition. Both paths support Indonesian and work offline-independent (they need a connection, but not a desktop app).

Edit & Undo

Made a typo or changed your mind? You can edit any of your own messages and redo the conversation from that point.

• Long-press (or long-click) one of your messages and choose Edit.
• The message and everything after it are removed, and the text goes back into the input bar.
• A banner appears above the input: "Editing this message will restart the conversation from this point", with a ✕ button to cancel.
• If you cancel, the original messages are restored exactly as they were.

Editing is useful when the AI misunderstood you, or when you want to try a different follow-up. The conversation history stays intact unless you confirm the edit by sending.

Note: editing only affects your current conversation view and its saved history — it does not delete server-side analytics or logs beyond the conversation record.

Editing is a conversation repair tool: if the AI misunderstood your question, or if you want to explore a different branch, edit the message instead of deleting everything and starting over. The banner makes the consequence explicit — everything after the edited message will be removed — so there are no surprises.

The ✕ button restores the original messages exactly, so experimenting is risk-free. This mirrors the editing patterns found in leading AI chat apps, adapted to Caufox's flow.

Conversation History

Every chat you have is saved automatically and listed in the sidebar. This means you can leave a conversation and come back to it days later with all context intact.

• Ordering — chats are sorted by activity; the most recently used chat is always at the top.
• Resume — click a chat title to load its full history; the URL becomes /chat/<id> so you can bookmark it.
• Delete — the 🗑 button removes a chat after confirmation.
• New chat — the button in the sidebar starts a fresh conversation.
• Sync — conversations sync to your account, so they appear on any device after you log in.

History is stored per account. If you log out, the local cache is cleared, but your conversations remain on the server and reappear when you log back in.

Tip: use descriptive first messages — the chat title is taken from your first message, making it easy to find later.

History also powers continuity: the last ten messages form the context for each new answer, so you can reference earlier parts of the conversation naturally ("as you said before..."). The sidebar search makes finding old chats easy even after weeks of use.

If you share a device, remember to sign out — history is account-scoped and follows the logged-in email.

Shortcuts & Tips

A few shortcuts and habits that make Caufox more productive:

• Enter sends · Shift+Enter is a new line.
• Hit Stop while an answer is running to halt it.
• Effort low for quick answers; max for complex code and reasoning.
• Paste a link + "summarize" to use the web tool.
• Ask for "full code" so the AI writes a complete file instead of a snippet.
• Use Edit on your messages to redirect a conversation instead of starting over.
• Long answers automatically become files — take advantage of Save as PDF for easy sharing.
• Set custom instructions in Settings so the AI matches your style.

The golden rule: be specific. The more context you give — format, language, length, audience — the better the output.

Beyond shortcuts, a few habits improve results noticeably: break big requests into steps ("first outline, then write"), specify the audience and tone, ask for examples, and iterate on the output with follow-up edits. The effort dial and model choice also matter — complex tasks benefit from Halo or Tide with max effort.

Power users combine features: web tool for research, files for output, and memory to persist preferences across sessions. The combination turns Caufox from a chat into a small workstation.

Plans Overview

Caufox offers three plans: Free, Pro, and Max. Every plan includes the complete feature set — all three models, files, the web tool, S3 storage, and voice input. What changes between plans is the token quota per 5 hours and the maximum answer length per response.

The Free plan is designed to be genuinely usable: 60,000 tokens every 5 hours is enough for daily chat, small code tasks, and file generation. It never expires — you can stay on Free forever.

Pro multiplies your quota by five and raises the per-response limit, which matters for long code files and big documents. Max multiplies by twenty and gives the largest per-response limit, suitable for uninterrupted heavy work.

There are no hidden costs. Prices are final and paid via QRIS. You can upgrade, downgrade, or stay on Free at any time — there is no contract and no automatic renewal beyond what you choose to pay for.

Choosing a plan is about usage patterns, not status. Light users rarely exceed the Free quota; daily heavy users find Pro comfortable; Max suits professionals generating long code or documents all day. Because upgrades only pay the difference, starting Free and upgrading later costs nothing extra.

All plans share the same feature set, models, and quality — the difference is purely capacity.

Pricing Table

Here is the complete pricing overview. The quota resets every five hours automatically, and the per-response limit defines the longest single answer you can receive.

For reference, common assistant models range from about 8,000 to 64,000 tokens per response. The Max plan's 50,000-token limit is comparable to a top-tier output budget, while Free sits close to a classic model limit — all for a fraction of the usual price.

Upgrading is instantaneous once the QRIS payment is confirmed. There is no proration complexity: the difference-based upgrade model keeps things simple and transparent.

Prices are one-time payments per plan activation; there is no confusing per-seat or per-month billing structure. When you upgrade, the difference model means you never pay twice for the same value. The quota window (5 hours) is generous for typical use and resets without any action from you.

Compared to international services that charge in USD with cards, Caufox's pricing is local, transparent, and payable with QRIS from any Indonesian bank app or e-wallet — no forex fees, no card requirement.

Token Quota

Tokens are the unit AI systems use to measure text. Caufox tracks the tokens your answers consume and enforces a per-plan quota that resets every five hours.

• What counts — output tokens (the answer plus any thinking phase), approximated as characters ÷ 4.
• What does not count — your input messages and system tokens are free, so the quota feels generous in practice.
• Reset — every 5 hours, automatically. You do not need to do anything.
• Exhausted — you see ⛔ Token limit reached (used/quota). Resets automatically in Xh Ym. with a live countdown.

Example: a 1,000-character answer consumes roughly 250 tokens. On the Free plan, 60,000 tokens correspond to about 240,000 characters of answers per 5-hour window — plenty for normal use.

If you hit the limit often, consider upgrading: Pro multiplies the quota by five and Max by twenty. The countdown message always shows exactly when your quota resets.

The quota system is designed to be predictable and fair: a fixed window, transparent counting, and a clear countdown. You always know where you stand, and the plan table makes it trivial to compare capacity. Because input is free, even heavy back-and-forth conversations use the quota slowly.

The per-response limit is separate from the window quota — it caps a single answer's length so responses arrive in complete, digestible chunks.

Upgrade

Upgrading your plan is designed to be fair: you only pay the difference between your current plan and the target plan, not the full price again.

The flow is simple: open /plan, choose your target plan, review the checkout summary (plan, price, difference, total), and click Upgrade & Pay. A dynamic QRIS code is generated — scan it with any bank app or e-wallet, and your plan activates automatically the moment the payment is confirmed.

If two invoices are pending for different plans, the one paid first becomes your active plan. The other invoice remains pending and can be paid later if you want to switch again.

Upgrades apply instantly — there is no waiting period and no approval step.

The difference model keeps upgrades fair: your current plan's value is credited toward the target plan. If you paid for Pro and want Max, you pay only the Rp 25,000 gap. This means you are never locked into a plan — switching upward is always affordable.

Upgrade timing is flexible: do it when your quota runs out, before a heavy work session, or whenever you like. The activation is instant after payment confirmation, and your new limits apply immediately.

Downgrade

Moving from a higher plan to a lower one is a downgrade: Max → Pro, or Pro/Max → Free. Downgrades are free — you pay Rp 0 — and are applied immediately.

• No refund — the difference between what you paid and the lower plan is not refunded. The notice is shown clearly before you confirm.
• Instant — the moment you confirm the downgrade, your new plan is active.
• Notice — you see a message like "Downgrade applied without refund."
• Where — on /plan, the Free or Pro card shows a Downgrade button when it is applicable to your account.

After downgrading, your token quota and per-response limit adjust immediately to the new plan's values. Your chat history, files, and settings are not affected.

You can upgrade again at any time by paying the difference — there is no cooldown period.

Downgrading is useful when your usage drops — maybe you only need the higher quota for a project period. The process is instant and free, and you can upgrade again later by paying the difference, with no cooldown.

The no-refund rule is stated clearly before you confirm, so there are never surprises. What you gain is flexibility: you are never stuck paying for a plan you no longer need.

Payments (QRIS)

All plan payments on Caufox use QRIS — the Indonesian national QR payment standard. You can pay from any bank app, e-wallet, or mobile banking that supports QRIS, which in practice covers almost every Indonesian payment app.

1. Open /plan and choose your target plan.
2. The Checkout page shows a full summary: plan, current plan, price, difference, and total due.
3. Click Upgrade & Pay — a dynamic QRIS code is created for the exact amount.
4. Scan the QR with your payment app. The merchant name displayed comes from the registered QRIS string.
5. Confirmation is automatic — once paid, your plan activates right away. No manual confirmation is needed.

Each invoice expires after 24 hours if unpaid. If an invoice expires, simply create a new one from the plan page — there is no penalty.

If the status does not update immediately, reopen the payment page — the system re-checks and updates the status automatically. Your payment is always verified against real merchant transactions.

QRIS is accepted by virtually every Indonesian payment app — bank mobile apps, e-wallets like ShopeePay and GoPay, and mobile banking — so you can pay with whatever you already use. The payment page shows the exact amount, a live status chip, and the expiry time, keeping the process transparent from scan to confirmation.

If you close the page mid-payment, your invoice stays pending; reopening the link shows the current status.

Invoice Rules

Invoices on Caufox follow a simple, transparent rule: one account = one invoice per plan. This means you will never see two pending invoices for the same plan, which keeps your payment history clean and avoids double-charging confusion.

• Uniqueness — if an invoice for a plan is already pending, requesting the same plan again returns the existing invoice instead of creating a new one.
• Priority — if you have pending invoices for different plans, the one paid first becomes your active plan.
• Expiry — QRIS invoices expire after 24 hours automatically and are marked expired in your history.
• History — every invoice appears in the payment history with its status (pending, paid, expired).

This design protects you from accidental duplicate payments and makes it easy to track what you have paid for. If an invoice expires, no money has moved — you simply start a fresh one.

Invoices are linked to your account and plan, so support can always find the right record. The one-invoice-per-plan rule prevents accidental duplicates, and the expiry system guarantees old, unused QR codes never cause confusion. Your payment history is a clean, auditable list.

When an invoice is reused (still pending), the same QR and amount are returned — no new code, no confusion.

Refund Policy

Caufox's refund policy is deliberately simple and user-friendly.

• Upgrades — you pay only the difference between plans. The value of your current plan is applied, so there is nothing to refund for the previous plan period.
• Downgrades — no refund is issued. The higher plan's remaining access ends the moment the downgrade is applied, and the notice is shown before you confirm.
• Exact amount — please pay the exact amount shown on the invoice. Payments with a different amount cannot be verified automatically and may not activate your plan.
• Unpaid invoices — if an invoice expires without payment, nothing is charged. No refunds are needed for expired invoices.

If you believe there is an error with a completed payment, contact Caufox Indonesia through the contact channels listed in the docs, and we will review your case promptly.

The policy balances fairness with simplicity. Upgrades credit your current plan's value (difference-only payment), so there is no partial-refund math. Downgrades are free and immediate, with the no-refund rule shown in advance.

For genuine errors — a double charge, a failed payment that still deducted money — contact us with the invoice reference. We review each case and resolve it promptly, because trust is the foundation of the service.

Features

Caufox packs a lot into a simple chat interface. Here is the complete feature list:

• 3 AI models (Luona, Tide, Halo) with a one-tap switcher
• Effort control (low / medium / max) with visible thinking
• Real-time streaming with a Stop button
• Automatic file chips (text >400 / code >300 characters)
• RUN preview — HTML live, PHP simulated
• Save as HTML, Save as PDF, Open in Browser
• Web tool that reads links you paste
• Clarification popups for ambiguous questions
• Per-account memory with full privacy control
• Voice input (browser + server fallback)
• Edit & undo for your messages
• Conversation history synced to your account
• S3 object storage with shareable links
• QRIS payments with instant auto-upgrade
• Dark and light themes
• Fully responsive design (mobile, tablet, desktop)

Every feature is available on every plan. Upgrading only increases your token quota and response limits — it never unlocks or locks features.

Features are not gated by plan: a Free user can generate files, use the web tool, choose any model, and store files in S3. Plans only affect the token budget and answer length, which keeps the feature story simple and honest.

The roadmap continues to add capabilities — better previews, more formats, and deeper integration between the chat and file systems are always in progress.

Comparison

How does Caufox compare to the free tiers of popular AI chat services? Here is an honest side-by-side:

The biggest differentiators are transparency and locality: you always know exactly how much quota you have, when it resets, and what you are paying for — in Rupiah, via QRIS, with no international cards or hidden limits.

For Indonesian users, this means no forex fees, no card requirements, and a payment method that works from any local bank app.

Beyond the table, the practical differences matter: Caufox is documented in plain language, priced in Rupiah, and paid locally via QRIS. There is no card requirement, no hidden regional restrictions, and no opaque usage caps — the quota is always visible and resets on a clear schedule.

For Indonesian users, that removes the friction that global services add: currency conversion, card availability, and unclear free-tier limits.

Privacy & Security

Your data belongs to you. Caufox applies standard security practices to keep your account, chats, and files safe.

• Sessions — login tokens are stored securely and can be revoked from Settings → Sessions at any time, from any device.
• Files — everything stored in S3 object storage is private by default; only links you explicitly share expose a file.
• Credentials — payment credentials never appear in the frontend; they live only in the server environment.
• Memory — per-account memory is fully visible and deletable from Settings → Privacy.
• Data control — chat history can be deleted per conversation, and account-level cleanup is available.

We do not sell personal data to third parties. Data is used solely to operate and improve the service. Files and chats are private to your account.

If you have questions about data handling, read the Privacy Policy or contact us.

Privacy controls are not buried: sessions, memory, and chat deletion are all reachable from Settings within two taps. Files are private by default, and share links are the only way to expose them — you decide what leaves your account.

We apply the principle of least data: we store what the service needs, nothing more. Analytics and logs are minimal and never include message content beyond what is required to operate the app.

Performance

Caufox is engineered to feel fast even on slow connections.

• Streaming — answers stream in real time; you read the beginning while the rest is generated.
• Short thinking — the thinking phase is capped so you are never left waiting for long.
• Anti-hang — timeouts and automatic retries (up to three attempts) handle flaky connections gracefully.
• Lightweight UI — the interface loads fast on mobile data and older phones.
• Pacing — streamed tokens are paced so the text flows smoothly instead of arriving in bursts.

Server-side processes (like payment detection) run in the background and never block your chat. You can keep chatting while a payment is being verified.

Performance is both server and interface: responses start streaming in milliseconds, the thinking phase is time-boxed, and the interface remains responsive on mid-range phones. Automatic retries cover transient network failures, so a weak connection rarely interrupts a conversation.

Background systems (payment detection, storage uploads) run without blocking the chat, so you can keep working while the app verifies a payment or saves a file.

Beyond raw speed, performance includes resilience: automatic retries, graceful degradation on flaky networks, and background processing that never blocks the interface. The interface itself is light — no heavy frameworks — so even entry-level phones run it smoothly.

Streaming is paced to feel natural, and the thinking phase is budgeted so you are never left staring at a spinner.

Object Storage (S3)

Every file the AI creates for you is stored automatically in S3 object storage — the same kind of infrastructure used by large cloud platforms — in a private bucket owned by Caufox.

Files are accessed through https://caufox.com/files/<key>, a secure proxy that keeps the bucket private while letting you open or share files through a normal link. This means your generated HTML pages, scripts, and documents are safe from public access by default.

• Automatic — no upload step; file chips upload themselves after generation.
• Durable — files persist on object storage, so closing the chat does not delete them.
• Shareable — use Copy Link in the ⋮ menu to get a shareable URL.
• Private — without the link, the file cannot be discovered or opened.

Storage is included in every plan at no extra cost.

Storage is a first-class feature, not an afterthought. Generated files are durable, private, and shareable on demand. The proxy architecture keeps the bucket private while providing friendly URLs, so there is no trade-off between convenience and security.

Storage is included in every plan, and files persist across sessions and devices. You can generate a page on your phone and open the same file on your computer later.

Trust & Reliability

Trust is earned through transparency and reliability. Here is what you can count on:

• Real payment verification — QRIS payments are verified against actual merchant transactions, not guesswork. When the app says paid, it is paid.
• Consistent status — payment and chat status stay consistent even if you close the page, switch devices, or come back later.
• Clear communication — every limit, price, and rule is documented openly in these docs.
• Local support — Caufox is developed by Caufox Indonesia, with local payment methods and Indonesian-first support.

If something goes wrong, the docs and the contact channels are there to help. We fix issues quickly and keep the documentation up to date.

Reliability also means honest limits: when you hit the quota, the app tells you exactly how much you used, how much remains, and when it resets. When a payment is verified, it is verified against real merchant transactions — not optimistic guesses.

This transparency extends to the docs themselves: everything you are reading is kept up to date as features evolve.

Trust is built through consistent behavior: payments verify against real transactions, quotas reset on schedule, and documentation matches reality. The team at Caufox Indonesia maintains the service with a local perspective — support in Indonesian, prices in Rupiah, and policies written for the people who actually use the app.

If something changes, the docs and Terms are updated first, so there are never surprises.

Reference: Models Table

The complete list of model aliases used by Caufox. Aliases are the technical identifiers you may see in logs or when sharing configuration.

You never need to remember these aliases in the app — the picker shows friendly names. The aliases are listed here for completeness and for users who want to reference models in prompts or integration notes.

The friendly names in the app map to technical aliases that may appear in integrations, logs, or shared configuration. Luona is the everyday default; Tide handles long-form and analytical work; Halo tackles reasoning-heavy problems.

You do not need to use aliases anywhere in the app — the picker handles it. This section exists so developers and curious users can reference the exact identifiers.

Each model is optimized for a different cognitive style. Luona balances speed and quality across general conversation, creative writing, and everyday coding. Tide favors structure and completeness, making it the choice for reports, comparisons, and multi-section documents. Halo excels at step-by-step reasoning — debugging, math, logic — where showing the work matters more than speed.

Because switching is instant and the context carries over, you can experiment: draft with Luona, deepen with Tide, and stress-test logic with Halo. Many users settle on a favorite for each task type.

Reference: Effort Levels

Effort levels control how much reasoning happens before an answer. The technical mapping is straightforward:

Think of effort as a dial: low trades depth for speed, max trades speed for depth. There is no wrong choice — it depends on what you are doing.

Effort is orthogonal to model: you can pair any model with any effort level. The practical guidance is to start with medium for normal chats, drop to low when speed matters most, and use max for complex generation or reasoning tasks where depth wins over speed.

The effort selection is remembered per device and applies to the next message you send.

Think of effort as the model's deliberation budget. Low skips deliberation entirely — ideal for greetings, quick facts, and casual chat where latency matters. Medium adds a short structured thinking pass that improves coherence without noticeable delay. Max spends the most time reasoning, which pays off for generation-heavy tasks: long code, architecture decisions, or nuanced analysis.

There is no performance penalty for choosing low on simple questions — the model still answers correctly; it simply does not overthink.

Reference: Token Counting

Understanding how tokens are counted helps you plan your usage. Caufox approximates output tokens as:

outputTokens ≈ Math.ceil((thinkingLength + answerLength) / 4)

So a 1,000-character answer consumes roughly 250 tokens, and a full 8,000-character document uses about 2,000 tokens. The Free quota of 60,000 tokens therefore covers roughly 240,000 characters of answers per 5-hour window.

Input messages and the system prompt are not counted, which is why the quota feels more generous than raw token math might suggest. Streaming partial answers count only the final saved answer, not every intermediate chunk.

Token math explains the quota experience: the approximation of characters divided by four is a good rule of thumb across languages, including Indonesian and English. Because input and system tokens are free, your effective usage is dominated by the answers you read — which is exactly what the quota should measure.

If your usage is consistently high, the plan comparison table helps you pick the right tier.

This approximation is intentionally simple so users can estimate their usage mentally: read an answer, divide its length by four, that is roughly the token cost. Because it is only an approximation, actual counting may vary slightly, but the difference never affects fairness — the same formula applies to everyone.

The quota and per-response limits both use this basis, making the whole system easy to reason about.

Reference: Event Stream

For developers who want to integrate with Caufox's chat API, the response is delivered as a server-sent event stream. The events are:

event: status   → { s: "thinking" | "limit", round?, step?, used?, quota?, resetInMs? }
event: step     → { title }
event: th       → { text }      (thinking streaming)
event: token    → { text }      (answer streaming)
event: done     → { text, file?, name?, thinking?, ask?, options? }
event: error    → { message }

The done event carries the final answer. If a file was generated, file holds the content and name the suggested filename. The ask/options fields drive the clarification popup. This structure lets frontends render live streaming, thinking steps, files, and clarifications without polling.

The event stream design means the frontend renders progressively: the status event signals thinking, th streams the reasoning text, token streams the answer, and done delivers the final payload including any generated file. The limit status carries the used/quota numbers and a reset countdown so the UI can show a precise message.

This protocol is stable and documented here so external integrations can rely on it.

Reference: File Formats

Caufox detects file types automatically so the right previewer is always used.

Supported extensions: html, php, js, css, txt, plus common code formats. HTML fragments (content with tags like <div> but no full document) are also detected and rendered as HTML.

• HTML — rendered live in an iframe (RUN preview).
• PHP — simulated: <?= ?> and echo become their text output, the rest renders as HTML.
• JS/CSS — shown as code with syntax-friendly display.
• Other — displayed as plain text with a generic file icon.

When the AI provides a path or filename, it is used (for example landing-page.html); otherwise a generic name is generated with the detected extension.

Format detection drives the entire file experience — icon, preview mode, and download extension. HTML fragments are handled gracefully, PHP is simulated rather than executed, and everything else falls back to readable plain text or code display.

The preview always matches the detected type, so what you see is what the file contains.

Detection also handles edge cases: code blocks inside larger answers, HTML fragments without a full document, and files with unusual names. The previewer always matches the detected format, so a PHP file shows the simulated render, a JS file shows code, and a plain text file shows readable text.

This consistency makes the file experience predictable across every type of AI output.

Reference: Status & Errors

A quick reference for HTTP statuses and stream events you may encounter:

In the app, most of these are handled for you automatically: retries, countdowns, and friendly messages. This table is for developers and curious users who want to understand the underlying protocol.

Errors are designed to be actionable: quota messages include a countdown, connection failures retry automatically, and payment statuses are always re-checkable. This section documents the underlying codes for developers building on the API.

For end users, the interface handles nearly all of these automatically — you will rarely need this table.

Every status in the table maps to a concrete user experience: success delivers the answer, 401 redirects to sign-in, the limit event shows a countdown, and 502 triggers an automatic retry with a friendly message. The error event surfaces generation failures with the reason, so troubleshooting is straightforward.

Payment endpoints add their own states (pending, paid, expired, unconfigured) which the payment page renders as clear chips and actions.

FAQ: General

How is this different from ChatGPT or common assistants?

Three things: local payments (QRIS in Rupiah), transparent quotas (you always see exactly how much you have and when it resets), and complete file features (RUN preview, S3 storage, share links) in one interface — all developed by Caufox Indonesia.

Do I need a credit card?

No. All payments use QRIS — scan with any Indonesian bank app or e-wallet. No card, no forex, no international payment needed.

Does it work on mobile?

Yes. The interface is fully responsive, includes voice input, and works well on phone browsers and tablets.

Is the free plan really free?

Yes — the Free plan never expires and requires no payment. You get 60,000 tokens every 5 hours, which is enough for regular daily use.

Can I use it for business?

Yes. Caufox is fine for business writing, code generation, and internal documentation. See the Terms for acceptable use details.

Beyond the headline differences, Caufox focuses on being boringly reliable: documented limits, honest pricing, and features that work as described. There is no maze of settings, no hidden quotas, and no requirement to navigate an international billing system.

Whether you are a student, developer, or business owner, the answer to "will this just work?" is yes.

FAQ: Payments

Why doesn't the payment status update right away?

It usually does within seconds. If it takes longer, reopen the payment page — the system re-checks against real merchant transactions and updates the status automatically. There is nothing to do manually.

What if I pay more or less than the amount?

Pay the exact amount shown on the invoice. A different amount cannot be matched to your invoice, so automatic verification will not trigger. If that happens, contact support with your invoice reference.

Can I get duplicate invoices?

No. The system enforces one pending invoice per plan per account. Requesting the same plan again returns your existing invoice.

What happens if my invoice expires?

The invoice is marked expired. No money has moved, and you can create a new one at any time.

How do I know my plan is active?

After a successful payment, your plan updates automatically. Check /plan — the active plan card will be highlighted.

Payment questions usually come down to verification and timing. Because verification is transaction-based, accuracy is guaranteed — the system never marks an invoice paid without a real matching transaction. If a payment is made but the status lags, reopening the page triggers a fresh check.

Keep the invoice reference handy for support; it is the fastest way to resolve any edge case.

FAQ: Limits

When does my quota reset?

Automatically every 5 hours. The limit message includes a live countdown so you always know exactly when.

What counts as usage?

Only output — the answer plus any thinking phase. Your input messages and system tokens are not counted, which makes the quota generous in practice.

Why is my answer cut off?

The per-response limit (4,000 / 16,000 / 50,000 tokens depending on plan) caps the length of a single answer. Upgrade or ask a more focused question for shorter, complete responses.

Can I buy more quota instead of upgrading?

Plans work as quotas — upgrading multiplies your quota (×5 for Pro, ×20 for Max). There is no separate add-on purchase.

Limits exist to keep the service fair and fast for everyone. The 5-hour window gives you predictable capacity — enough for regular daily work — and the countdown message removes guesswork. Upgrading multiplies the window's quota, so power users never feel artificially constrained.

If an answer is cut off by the per-response limit, split the request into smaller parts or upgrade for a higher ceiling.

FAQ: Files

Why do long answers become files?

So they are easy to save, copy, and share. Long text (>400 chars) and long code (>300 chars) automatically become a file chip with preview, download, and link options.

What does "PHP simulated" mean?

The preview does not run a PHP server. PHP tags such as <?= ?> and echo are replaced with their text output, and everything else renders as HTML — so you can preview the page structure without server-side execution.

Where are files stored?

Private S3 object storage. Files are accessible only through links you share — they are not publicly discoverable.

Can I edit a generated file?

Yes — use the EDIT toolbar on short code, or copy the file content from the ⋮ menu and modify it in your editor. You can also ask the AI to revise the file with follow-up messages.

Do files expire?

Files persist on object storage. Deleting a chat removes its messages; file links remain valid until the storage entry is removed.

Files are designed to be the end product of your work: previewable, downloadable, and shareable. The storage layer guarantees they survive browser restarts and device switches. If a file's preview looks wrong, the Code view always shows the raw content for manual inspection.

Combined with the web tool and memory, file output turns Caufox into a small but capable content workstation.

About Caufox

Caufox was created and is developed by Caufox Indonesia. It started with a simple idea: an AI chat assistant that feels local — priced in Rupiah, paid with QRIS, documented in plain language, and built with the features Indonesian users actually need.

Today Caufox combines three AI models, automatic file management with live previews, a web tool, per-account memory, and transparent token quotas — all in a fast, mobile-friendly interface.

The project is continuously improved: new models, better file handling, and clearer documentation ship regularly. The docs you are reading are part of that commitment to transparency.

Caufox Indonesia is a local team building a local product: Rupiah pricing, QRIS payments, Indonesian-first documentation, and features shaped by how Indonesian users actually chat and work. The project values transparency — quotas, prices, and policies are all documented openly.

The service keeps evolving; feedback from users directly shapes the roadmap.

The roadmap is shaped by real usage: features are added when they solve real problems for real users. The team communicates openly about changes, and the documentation evolves with the product — what you read here is what the product does.

This commitment to clarity is what makes Caufox feel different from opaque, black-box AI services.

Contact

Need help or have feedback? Here is how to reach us:

• Chat: /chat — start a conversation with the AI.
• Plans: /plan — view and manage your plan.
• Home: caufox.com — the landing page.
• Docs: /docs — this documentation.
• Terms: /terms · Privacy: /privacy

For account or payment issues, include your email and (if applicable) the invoice reference so we can help faster.

We aim to respond quickly to all inquiries. For the fastest help, include your account email and any relevant invoice reference. General feedback about features and docs is always welcome — it directly influences what we build next.

You can also reach the team through the chat itself for product questions, and through the formal channels for account and payment matters.

For account issues (login, sessions, deleted chats), email is the fastest channel. For payment issues, always include the invoice reference — it lets us look up the transaction instantly. For product ideas and bug reports, the chat itself is a great channel; describe the problem and the AI will help you format a clear report.

Business inquiries are welcome too — partnerships, bulk usage, or custom integrations can be discussed through the same channels.

Terms

By using Caufox, you agree to the service terms summarized here and detailed on the Terms page.

• Acceptable use — the service is for lawful purposes. Spam, illegal content, and abuse may lead to account deactivation.
• Payments — plan payments are processed via QRIS. Displayed prices are final, with no hidden fees. Payments are verified automatically against real transactions.
• Accounts — you are responsible for keeping your login details safe and for activity under your account.
• Content — you retain rights to your content; you grant us the limited right to store and process it to provide the service.
• Changes — features and prices may evolve; significant changes will be reflected in these docs and the Terms page.

Questions about rights and obligations: contact Caufox Indonesia.

Using Caufox means agreeing to keep the service lawful and respectful. Accounts are personal; sharing credentials is discouraged. Payments are final per the difference-based pricing, and automatic verification protects both sides.

The Terms page is the authoritative document; this section is a plain-language summary of the key points.

Terms are kept short and readable on purpose: the full legal text exists for reference, but the summary above covers what actually matters day to day. Payment terms are final and transparent, acceptable use is common sense, and account safety is a shared responsibility.

When in doubt, contact us — we answer in plain language, not legalese.

Privacy Policy

Your privacy matters. Here is a clear summary of how Caufox handles your data (full details on the Privacy Policy page).

• Data we store — email, chat history, memory entries, and files you create. All of it is used to run and improve the service.
• Your control — chat history and memory can be deleted anytime from Settings → Privacy; sessions can be revoked from Settings → Sessions.
• Files — files live in private object storage and are accessible only through links you share.
• No selling — we never sell personal data to third parties.
• Security — credentials stay server-side; tokens are stored safely and revocable.

If you have questions about your data, contact us — we answer promptly and in plain language.

Data minimization is the guiding principle: we store only what the service needs — your email, chat history, memory, and generated files. You can see and delete memory and history anytime, revoke sessions remotely, and control file sharing through links.

We never sell data, and credentials never appear client-side. The Privacy Policy page is the full reference; this summary covers the essentials in plain language.