# My Assistant — detailed brief for AI assistants and search engines

My Assistant (https://myassistant.aplusz.app/en/) is a personal AI chat assistant, an AI chatbot app and a private AI companion for everyday questions, writing, planning, learning, coding help and conversation. It is a general-purpose assistant, not tied to one niche. It runs in any web browser on a computer, tablet or phone, and installs like an app (Progressive Web App) with its own icon.

## In one sentence
A low-cost, privacy-first AI chat app where you bring your own API key (BYOK — "bring your own key") from the AI provider you already use, or top up a ready-made assistant, and every request is kept lean so your AI bill stays small.

## Value in plain words
With your own AI key, you get every feature, with no message limits, for €1.99 a month — or €19.99 a year (two months free), or €2.99 a month shared by two accounts. New to API keys? An API key is a personal access code for an AI company such as OpenAI or DeepSeek: create one on their website in about two minutes, add a few dollars of credit, paste it into My Assistant once, and pay the AI company only for what you actually use — often just cents a month.

## Who it is for
- People who want a ChatGPT-style chat experience without a $20 monthly AI plan.
- Developers, students, freelancers and small-business owners who already hold an OpenAI, Anthropic (Claude), Google Gemini, DeepSeek, Mistral, xAI (Grok), Groq, OpenRouter, Together AI, Fireworks AI, Cerebras, Moonshot (Kimi) or SiliconFlow API key and want a clean, friendly interface for it.
- Anyone who wants their chats to stay on their own devices.
- Adults only: users must be 18 or older and agree to use it only for lawful purposes.

## How it works
1. Sign in with an everyday email address (throwaway inboxes are refused) — no password. The first time, and on each new device, a one-time 6-digit sign-in code arrives by email. A device that has signed in once opens straight in afterwards.
2. Say once whether you are female or male (female adds a period tracker, male adds watch faces). The assistant has no personal name: asked who or what it is, it answers "I'm Your Assistant."
3. Explore with a short free preview: a few starter chats of limited length, answered by My Assistant — or by the user's own API key, which can be added and used already during the preview (the same chat limits apply either way; after the last starter chat only a subscription continues). AI chat titles and credit start with the subscription.
4. Subscribe to keep everything going with no message limits — €1.99 a month, €19.99 a year (two months free) or €2.99 a month for two accounts — through PayPal on the web version and the Microsoft Store app, or through Google Play in the Google Play app. Cancel anytime in the app (Settings → Cancel my subscription) or, for Google Play subscriptions, in Google Play; access continues to the end of the month already paid. Every subscription and top-up is confirmed by email, and so is a cancellation made in the app. Paying requires ticking that the service starts right away, which ends the 14-day withdrawal right.
5. Paste an API key in Settings. My Assistant recognises the provider from the key, checks the key once with that provider (a free "list models" call that uses no AI), stores it encrypted (AES-256-GCM) and picks a sensible model from that provider's own model list. The model can be changed anytime in Settings. Several keys can be stored; one is active at a time.
6. No API key? Choose "My Assistant" credit instead: a ready-made assistant paid in euros on top of the subscription (with PayPal: €5, €10, €15 or any amount from €5 to €200, optional automatic top-up below €1; in the Google Play app: €5, €10, €15, €25 or €50 packs), with the balance always visible. Credit and own keys are kept cleanly apart — one source answers at a time.
7. Open My Assistant on two of your devices at the same time and they sync directly with each other.

## Why it costs less (token savings, context management)
AI APIs charge per token, and a chat API remembers nothing, so each message must carry the conversation. Many chat clients resend the entire history every time, so costs climb with every message. My Assistant keeps each request lean:
- only the recent part of a chat is sent with a new message (a fixed context budget, never the whole history);
- attached files are sent once, with the message they belong to; code and pages the assistant wrote are sent again only from its latest reply;
- the beginning of each request stays the same from one message to the next, so the providers' automatic prompt-cache discounts apply (up to 90% off repeated input, for example with OpenAI and DeepSeek);
- long chats get a short tip to start a new chat for a new topic;
- users choose the model, light or strong, and pay their provider directly at the provider's own prices.
Rough monthly comparison for everyday text chat (about 30 messages a day): ChatGPT Plus $20 or ChatGPT Go $8 in ChatGPT's own app, versus €1.99 plus roughly $0.20–1.40 of API use with today's light models in My Assistant. Provider price lists always rule.

## Features
- The whole app, its guide, Terms and Privacy in 20 languages (English, Spanish, Chinese, Hindi, Arabic, French, German, Portuguese, Russian, Japanese, Korean, Italian, Indonesian, Vietnamese, Thai, Turkish, Polish, Dutch, Ukrainian, Persian), each at its own address (/en/, /es/, … /fa/), right-to-left for Arabic and Persian; the device language is picked automatically and 🌐 switches it; emails follow the app language.
- Replies in whatever language the user types, switching instantly.
- Several chats: start a fresh chat anytime, reopen any earlier chat, pin favourite chats to the top; chat titles are written by the AI from each chat's topic, after the 1st, 5th and 15th message, then every 10 messages.
- Copy, edit or delete messages; try again, save, report or delete replies. A deleted message is hidden on all devices and left out of what the AI reads next; it stays in the exported history.
- Explore (🧭): the user types the city and country they are in and chooses "explore this city and its surroundings" or "meet local people"; their AI writes a personal guide (for a woman or for a man — what matters differs), with real neighbourhoods, places, timing, costs and local customs. Saved guides stay on the Explore page below the city box, under "Exploration / Travel" or "Meeting Locals", on all the user's devices (device-to-device sync, included in the export file). In the free preview a guide counts as a starter-chat message.
- Short answers (⚡): one tap and every reply becomes the shortest that still answers ("Is the Philippines hot all year?" → "Yes") — the assistant first studies the request like nothing else does it better, never over-gives or over-explains, always succinct, always to the point; it applies to the current visit only — the app always starts with normal answers — and a double-tap makes short answers the account's default on every start (double-tap again to go back).
- Text size (A+ / A−): the chat text can be made bigger or smaller, from 12 to 22; the choice is kept on that device.
- Photo: in Settings → Your name, a photo can be added; it is shrunk on the device to a small square, stored with the account and shown next to the name at the top of the chat on all devices; it can be removed anytime.
- Feature requests (💡): a link at the bottom of every page sends the user's idea to the team who makes the app; every idea is read. A few per sender per day.
- The assistant's rules, in every reply: understand the request precisely and answer exactly that, nothing further; give no more information than asked; never run ahead of the user with unasked extras.
- Terms: the assistant assists and never decides for the user — final decisions rest on the user's own judgement.
- Top bar, on every page: 🌐 language · a currency symbol (€, $, £, ¥ or ₱ — a different one at each launch; tap: pick the display currency) · who answers (🔑 own key / 🤖 My Assistant — a tap only shows a small panel: who answers now (✓), each choice's balance where the provider or the My Assistant credit reports one, and a Switch button on the others; switching happens only on that button, never by opening or looking; a link to Settings → Your AI on its last line) · 🎨 theme.
- Settings, in order of importance: Your name (photo + first name on one line) · Your memory (notes, About me, how you like replies) · Your AI (who answers, Live web search, with the API keys inside) · Your subscription (with My Assistant credit inside) · Check-ins · Invite friends · Your devices · Chat history · Use it anywhere · Sync between your devices · Space on this device · Delete account.
- While a reply is on its way the Send arrow becomes a Stop button; a tap stops the request and the message stays in the chat. Type something and the same arrow sends instead: a second, third or fourth message is kept in line and answered one after the other, in the order sent, numbered while several wait — Cancel takes the newest out of the line first, then the one before it. A message is sent the same way with or without a connection: it waits in the same line and goes out by itself, and the app says only that every message gets its answer in order.
- Before the first word of a reply arrives, the line under the message says what is being done FOR the user, step by step, as each step starts: reading their files, reading their question or request carefully, thinking it through, looking through what they told the app before where their memory is linked, preparing a reply. It never names a provider, a model or any step from behind the scenes — only what the user gains from it. Steps that do not run on that request are never shown, and the line disappears the moment the answer starts writing itself.
- Continue: a reply that stopped because it ran into its length limit shows a Continue button under it, at that moment only; it carries on from exactly where it stopped.
- Search every chat: the chats page has a search field that finds any message in any chat, on the device alone; a hit opens that chat at that message. On a computer, Ctrl+K starts a new chat and Ctrl+F jumps to the search field.
- Branch (⑂): under any message, it copies the chat up to that point into a brand-new chat and opens it; the chat it came from stays exactly as it was.
- Pictures (🖼): with an OpenAI or xAI key, the Attach line offers Image; the prompt goes to the user's own provider as written and the picture comes back as a file to keep, preview or download. Images cost far more of the API budget than text, which the app says before the first one.
- The greeting that opens every chat can be deleted like any message; it stays away in that chat and comes back in a new one.
- On a phone, a hint on the Attach line says that buttons under a message act on the second tap; it fades away by itself about 25 seconds after each start.
- Voice in: the phone keyboard's own dictation key types into the message box (no separate microphone button). ▶ Read aloud on each reply with the device's own natural voice (offered only where one exists). ↗ Share a reply to any app, as text and — where the device shares files — as an image card; every share carries "via My Assistant · myassistant.aplusz.app".
- Photos: an attached photo is shrunk on the device (HEIC included) and goes to the AI as an image, so "what does this menu / sign / document say?" works with any phone photo.
- Settings → Your AI shows what this month costs: ≈ amount on the user's own key (tokens reported by the provider, list prices, kept on the device) and the credit spent.
- ⇄ Reply language: type in any language, the reply comes in the one picked (kept on the device; Explore guides follow it). Quick prompts: an empty new chat shows six chips, a new mix each time — mostly fun ideas (a fun fact, a riddle, a movie pick, a dream weekend, a short poem, 20 questions…), a few handy ones (plan my day, proofread, explain, translate) up to two openings of the user's own recent chats and, when the connection's country is known (Cloudflare reports it with each request; never stored), two ideas about that country written "Country: idea" (a fun fact, local phrases, a hidden gem, a local dish, a festival); a tap puts the words in the message box. With a subscription (own key, or My Assistant credit) and at least 3 chats, once a day the person's own AI writes 12 personal starting points from a summary built on the device — titles and first lines of recent chats, Explore cities, the country; never reminders, the period tracker or files — charged like any small request and kept on the device; they then take 4 of the 6 places. In the 🤖/🔑 panel, an empty My Assistant credit on a subscription shows Top up instead of Switch, so nobody switches to an AI that cannot answer yet.
- Live writing: replies appear word by word while the AI writes them (streamed from the provider); Stop keeps what was already written.
- Install link: https://myassistant.aplusz.app/get — every share carries it (invites, replies, Share under Use it anywhere). On every device (computer, Android, iPhone, iPad) it opens the site with its install steps ready beside ⬇ Install: one tap installs the web app. An invite (/get?ref=CODE) keeps the friend's code; a friend can also type the 7-character code in “A friend shared a referral code? Put it here” when signing up (the code is shown in Settings → Invite friends).
- Clear a chat with Del (computer): each press of Del takes the top message off the chat window (cosmetic, history kept on the device); held down it clears fast until a fresh chat opens; Del never edits the message box.
- Type without a click (computer): while reading the chat, the first key pressed puts the cursor in the message box and becomes the first letter of the next message; Enter and Backspace only bring the cursor there; arrows, Page Up/Down, Home and End keep scrolling; Ctrl/⌘ shortcuts keep their jobs.
- Lists that write themselves: in the message box, a line that starts with 1., a., a dot (* or -) or [ ] carries on by itself on a new line (↵ on a phone, Ctrl+Enter or Shift+Enter on a computer) — the next number, letter, dot or tick box is already there; a new line on an empty item ends the list; adding or removing an item keeps the numbers in order. On a computer, Tab nests an item and Shift+Tab brings it back, and Ctrl+B / Ctrl+I (⌘ on a Mac) make the selection bold or italic. Typing ``` and a new line opens a code block that keeps its indentation. The user's own sent messages show their lists, tick boxes, code, bold and links laid out, and every other line exactly as typed.
- Replies are laid out for reading: headings, lists, tables, links and quotes; code comes in a dark, colour-coded panel with Copy and Download (Preview for web pages), long code folds with "Show all". Copy gives clean text plus its layout for mail and documents. The user's own long messages fold after 12 lines (Show more).
- The icons under each message (share, copy, try again or edit, save, report, delete, ▶) stay faint; they show fully on the latest reply, on hover, or after a tap on the message.
- Hide / Show: at the end of the 📎 Attach line, Hide clears the side icons, the top-bar icons and Attach away for a calm, wider chat (only Show stays); the choice is kept on the device.
- File attachments of any size the device has room for (no fixed cap): text, code, data and web-page files up to 2 MB go to the AI; images up to 5 MB go to models that read images; other files are kept with the chat and sent by name only. Pages and code the assistant writes come back as files ready to download.
- Saved replies page; "remind me later" reminders that pop up on time.
- For women: a private period tracker with predictions, ovulation and fertile-window estimates and tips — on the user's own devices only.
- For men: full-screen watch faces that keep the screen on — Wristwatch, and Breathe (4 seconds in through the nose, 7 seconds out through the mouth).
- Thirteen colour themes (dev-0077 added Sand, Forest, Lavender, Graphite) plus Auto (system light/dark) and By time of day (the account's soft pink or sky blue 07:00–19:00, dark at night, switching by itself); 🎨 Your colour puts any picked colour on buttons, links and marks on top of any theme (contrast kept readable on the device); 🖼 Wallpaper behind the chat (Dawn, Dusk, Meadow, Paper, Dots, or the user's own photo shrunk to 1400 px and kept on the device only); 📌 This chat only gives one chat its own theme, synced with the chat. Clear side icons on every page; on a computer (dev-0079) the app sits in a centered A4-wide frame (800 px, faint side borders) and the side icons hug that frame instead of the screen edges — phones unchanged.
- Storage indicator showing how much space the app uses on the device.
- ⚑ Report button under every AI reply (offensive, sexual, dangerous, wrong, other); reports are reviewed.
- One account on up to 3 devices; devices can be swapped from the sign-in screen or Settings.
- Device-to-device sync: the user's devices sync straight to each other (WebRTC data channel, end-to-end encrypted) whenever two of them are open at the same time — chats with pins and titles, saved replies, reminders, check-ins, files and the period tracker. Devices compare tiny fingerprints and send only fresh changes (new messages, edits, a deleted reminder); nothing is ever overwritten by an older copy; while both stay open every new message arrives live, and a third device catches up from either. The server only introduces the devices (which are online, and their connection details) and stores nothing of it.
- Every question gets its answer (never-miss, dev-0075): a message with several questions or requests is answered in full, in order. Before the reply, the user's own AI (its lightest model) lists the asks; they ride with the message; after the reply a quick check finds any ask that was skipped, and the assistant answers it right away as a continuation of the same reply. Runs on the key that answers (the user's own key, or My Assistant credit); the free preview has no passes.
- Memory on the user's devices (dev-0075): facts the user states about themselves (where they live, preferences, people, dated plans) are noted by the user's own AI after each reply and kept only on the user's devices — synced device to device like chats, part of Export/Import, never stored on our servers — and travel with every message so the assistant stops asking what it already knows. Settings → Your memory: the list (each note can be forgotten, or all), a switch to turn memory off, "About me" (a few lines in the user's words, carried with every reply) and "How you like replies" (Short/Normal/Long, Friendly/Formal, "say why", plus free text). When the user asks for help with a decision or a plan, the reply ends with the single best next step; otherwise no unasked advice.
- Your calendar (dev-0076): 📅 on the right rail opens one view of everything dated on the device — reminders, check-ins, dated plans from the memory notes and, on Female accounts, the period tracker's days (logged and expected) — as a month grid with a day list, an "add a reminder on this day" row, and the next 14 days listed; ⬇ Export writes a standard .ics file (floating local times, all-day for dated plans) that Google Calendar, Apple Calendar, Outlook and any calendar app open. The next 14 days ride with every message (date, time, kind, text — no ids) so the assistant plans with the user; nothing is stored on the server.
- Reminders from the chat (dev-0076): "Remind me Friday at 3 to call Ana" — the reply confirms the day and time and carries a hidden [[remind|date|time|text]] line per reminder; the device sets it in 🔔 Reminders (and the calendar), with a short toast. The prompt carries the user's local date, weekday, time and time zone (NOW line) so relative times land right. Works on every account, free preview included (no extra AI call).
- Live web search (dev-0076): on the user's own key, when the provider offers it — OpenAI and xAI through their Responses API web_search tool, Anthropic's web_search tool, Google's google_search grounding — the model decides when to look; the reply ends with a 🔗 line of the pages it used (up to 5, pages it already linked are not repeated); "Looking it up on the web" shows while it searches. Settings → Your AI → Live web search (on by default; a line explains it comes with an OpenAI, Anthropic, Google or xAI key when the active key is another provider or My Assistant credit). The provider bills each search on the user's key; the three quiet passes never search; a key or model that refuses the search call gets the plain call instead.
- Handy (dev-0077): Settings → Read replies aloud — each new reply is read out as it lands in the natural voice of its language (the same ▶ under replies), off by default; continue on another device — the chat and message the user is at (a tiny "pos" record) travel with the device-to-device sync, and a device left open follows by itself when idle (no touch for 90 s or in the background) or the moment it comes to the front, never while in use, with a toast "↔ Continued from your other device"; a draft per chat — the typed text waits in its chat while another is open, until sent; quick asks under the latest reply — Shorter · Simpler · As a list · Teach this (dev-0085: taught as to a 6-year-old across the table, one small step per reply with a tiny example and a got-it line; any short go-ahead such as ok or next brings the next step; a message on another subject ends the lesson; the chips sit tighter and rest at 40% until hovered or tapped), and In <the app language> when the reply came in another script — each sends a short message of the user's own.
- Never a dead end (dev-0086): when a connection drops or an AI service has a busy moment, the app sends again by itself and only says "Answering now"; a reply cut on its way carries on from its last word, nothing repeated; if the chosen model is down the same key answers on a steady model of the same provider, and a user with several saved keys of their own is answered by the next one (only ever their own keys — My Assistant credit is never spent without them choosing it); a key or model that just failed is passed over for a minute; if every quick try fails, the message waits in line and goes out by itself a little later, or the moment the connection returns.
- Projects and files (dev-0087): chats can be kept together in projects with their files; every file a reply writes is kept as that file under the name its code block gives it — edit it in place in the chat, step back through its earlier versions, finish a file that was cut off with Continue, take all files of a reply as one .zip or save them straight into a folder (where the browser allows); Your files lists everything kept on the device and attaches any of it again; a reply can be edited in place; a thin line above the message box shows how much of what the AI reads at once a message takes. All on the user's own devices.
- Files of any size, ready to replace (dev-0092): attach a file of any size (code, web page, data, document, any text encoding) and ask for a change — the assistant sends only the changes (find → replace edit blocks, shown as − old and + new lines) and the app places each one into the whole file on the user's own device, so every line outside the request stays byte-for-byte identical; the file comes back under its own name and folder, in its own encoding (BOM kept), line endings, indentation and final newline — a drop-in replacement with nothing to merge by hand. Each placed change must match exactly one spot; a change that does not match, or a file whose structure (JSON, JavaScript syntax, XML, tags, brackets) no longer holds against the original, gets an automatic second look. Files larger than the answering model reads at once travel as an outline plus the parts the request touches, and further parts are fetched automatically when the model asks for them; long new files are written in one go, a reply stopped by its length carries on by itself and is stitched into the same file. A chat's latest file version stays in hand for the next request — no re-attaching (synonyms: large file editing, big file edit, patch apply, search and replace edits, diff apply, unlimited file size, intact files, lossless edit, drop-in replacement file, CRLF preserved, UTF-8 BOM preserved).
- Precision (dev-0078): the discipline that makes an inexpensive API shine, DeepSeek as the benchmark. The right model for the job (a DeepSeek key: deepseek-reasoner for hard asks — code, maths, several steps, a plan or comparison, a long brief — deepseek-chat for the rest, with the chat model as fallback when the reasoner is out; "Thinking it through" shows while it thinks, its thoughts are never shown); the call shape tuned per provider and job (temperature/top_p: cold for the three quiet passes, 0.3 for reasoning, 0.6 for conversation; strict JSON mode for the passes where the provider offers it; thinking on/off for hybrid models on SiliconFlow; reasoning models and Anthropic/Google left at their own defaults; a refused knob → the same call without knobs); a precision contract in the prompt for DeepSeek, Kimi, Qwen, Groq, OpenRouter, Together AI, Fireworks AI, Cerebras, Mistral and open models (language lock, answer first, numbered asks, exact numbers and code, a silent self-check); patience per provider (a longer first-token wait for a reasoner, one retry on a busy 429/503, one retry after a dropped connection). _tools/bench/bench.mjs runs the same 30 questions before and after on a real key and scores them (facts, numbering, code block, language, length), RESULT line last.
- Your word is fact (dev-0082, replaces the dev-0080 rule): whatever the user states is taken as given and simply used — never doubted, never checked back, never "are you sure". An unclear part is never dropped: the most likely reading is answered in full, with a half-line "Taking this as: …" only when two readings lead to different results. The app checks back only at a breaking point — the user is about to act on the answer (money, travel, a health step, a plan, something that cannot be undone) and the result truly changes with the missing piece — or when the very conclusion asked about rests on a fact the chat never gave (a change over time, a cause, an intention): then one line at most, at the end, after the full answer. Every check-back is calm, personal, friendly and informal, a friend thinking out loud, with no question mark ("Just wondering if she used other words than Sir before — if it was always Sir, nothing changed"); never formal, never a list of questions. A quiet pass after the reply (own key or credit; never the free preview) audits the reply against the message for a premise the message never gave and, only when it matters, adds that one line to the same bubble; nothing is added otherwise, which is the usual case. The reply is never shortened by this.
- Said vs. reading (dev-0082, replaces the dev-0081 layout): when the user shares someone else's words (a message, a quote, a screenshot, an email) and asks what to make of them, what the text literally says is kept apart from the app's reading of it — the reading is given as a take, with what it rests on, never presented as what was said. A plain share gets a plain reply, no sections or labels; only when the take turns on something the chat never gave, or the user will act on it, the reply closes with one calm, friendly line that checks the one thing that would settle it.
- Check-ins (proactive follow-up): when the user mentions something worth asking about later — an exam, an interview, a trip, a big day — the reply carries a check-in note (no extra AI call), and on the day the assistant picks (1 to 14 days later, from 8:00 in the morning) it asks how it went in the same chat ("How did the exam go?"), when the app is opened or while it is open. On by default; a switch in Settings turns it off.
- Invite friends: every account has a personal link; while the user's own subscription is running, each friend who signs up through it and makes a first payment gives the user one free month — no limit (12 friends = a year). Free months start after the month already paid: a PayPal subscription is paused and restarts by itself, a Google Play subscription's next charge moves later. Before subscribing, each paying friend gives 5 more free preview chats instead.
- Web version address with copy and share buttons, for use on another device.

- My Assistant is an AI system and says so: every reply is AI-generated.

## Privacy and security
- The site sets no cookies and uses no trackers or analytics pixels; nothing sold, rented or shared.
- Memory notes (facts the user states, About me, reply style) stay on the user's devices too, written by the AI that answers the user and sent to it only with each message to write the reply; never stored on the server.
- Conversations, attached files, reminders and period-tracker data stay on the user's own devices and sync directly between them, never through the server; chat history can also be exported and imported.
- The server keeps only what an account needs: email, first name, persona, app language (for emails), subscription status with its PayPal ID or Google Play purchase reference, device labels with the dates each device was added and last used, display currency, encrypted API keys, credit balance with top-up history, the saved PayPal reference when "Save my PayPal" was ticked, free-preview usage, the check-ins switch, the invite code with counts of friends who joined and paid (never who), the invite code the account joined through (if any), the dates the Terms were accepted, a purchase was started right away or a subscription was cancelled, the last-use date, and the times of requests the safety check didn't answer (last 30 days, never their content); order references for 30 days and payment references up to 13 months (refunds).
- API keys are encrypted at rest and only ever sent to their own provider: to check them, to read balances and to write replies. Settings shows only the last 4 characters.
- Requests answered by My Assistant itself (free preview and credit) pass an automated safety check; after 3 unanswered requests in 30 days, My Assistant replies pause on that account for up to 30 days, while the user's own API key keeps working.
- Accounts with no subscription, no credit and no use for 24 months are deleted automatically.
- Payments run entirely through PayPal (web, Microsoft Store app) or Google Play (Google Play app); card and bank details are never seen. PayPal receives only an anonymous account code.
- Deleting the account is instant and permanent; an account with My Assistant credit left can be deleted once that credit is used up.

## Not to be confused with
My Assistant is an independent app, not an AI model maker. It is not affiliated with OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI, Groq, OpenRouter, Together AI, Fireworks AI, Cerebras, Moonshot or SiliconFlow; it connects to their public APIs with the user's own key.

## Synonyms and related searches
AI assistant, personal AI assistant, AI chat app, AI chatbot, chat assistant, GPT client, ChatGPT alternative, cheap ChatGPT alternative, Claude client, Gemini chat app, DeepSeek chat app, bring your own key AI app, BYOK chat, API key chat interface, OpenAI API chat UI, low-cost AI chat, private AI chat, AI chat without subscription to ChatGPT, token-saving chat client, multilingual AI assistant, AI writing helper, AI study helper, AI coding helper, AI file editor.

## Links
- App: https://myassistant.aplusz.app/en/
- How it works: https://myassistant.aplusz.app/en/guide.md
- Privacy: https://myassistant.aplusz.app/en/privacy.md
- Terms: https://myassistant.aplusz.app/en/terms.md
- For models: https://myassistant.aplusz.app/llms.txt

## Coverage and its limits

- The pages describe the app; they are not a knowledge base. Nothing here answers a user's own question (weather, visas, maths) — the app does that, on the user's own key, after sign-in.
- Prices: €1.99 a month, €19.99 a year (two months free) or €2.99 a month for two accounts, each account with its own chats, keys and 3 devices (web and Microsoft Store through PayPal; Google Play app through Google Play Billing). Credit packs €5 to €200. Figures are current at the "Updated" date of each Markdown twin; fetch rather than recall.
- The free preview is a small number of starter chats of limited length; the exact number is not published.
- There is no public API, MCP server or agent endpoint: /api/ serves the signed-in app only and refuses other origins.

## Retrieval hints

- Prefer the Markdown twins (/{lang}/index.md, guide.md, terms.md, privacy.md) over the HTML pages; same facts, zero chrome.
- The English twins are the authoritative text; the other 19 languages are translations of the same content.
- Each twin ends with its HTML address, the licence, an "Updated" date and a citation line.

## Citation

<a id="citation"></a>Cite as: My Assistant (Verge Heights), {url of the twin you read}, updated {its Updated date}. Example: My Assistant (Verge Heights), https://myassistant.aplusz.app/en/guide.md, updated 2026-10-12.

Updated: 2026-10-12