Lumi AI

An AI companion that knows you —
transparent, controllable, grounded.

Lumi isn’t a chatbot in the corner. It runs through capture, planning, analysis and growth, and appears as a quiet orb of light — when you see the spectrum, you know it’s Lumi speaking.

Four roles

Capture, arrange, analyze, encourage.

  1. 01

    Capture assistant

    Tasks, events, spending and feelings in one sentence — split and filed.

    “Buy milk tomorrow, report Friday, lunch 12”

  2. 02

    Life secretary

    Morning briefings, evening reviews, and Plan My Day — finding time for tasks between meetings.

    “Put this week’s study time on my calendar”

  3. 03

    Analyst

    Ask anything about your life and get charts, tables and sources you can open.

    “How much did I spend on food this month?”

  4. 04

    Growth coach

    Break down life goals, build growth plans, review them regularly — and remind you why when you want to quit.

    “I want to join a global company next year”

How Lumi understands you

Not a bigger model —
better context about you.

Models are converging. What decides how well an AI knows you is how much of your life it can see. Within the permissions you grant, Lumi understands you through:

  1. 1

    A life snapshot

    A short daily summary of who you are, what you’re working toward and how you’ve been — built only from modules you let Lumi read.

  2. 2

    Memory you manage

    Important things from chats, journals and planning interviews — each with its source, editable and deletable.

  3. 3

    Permission-aware retrieval

    Only the most relevant records are used; private content and excluded entries are never retrieved.

  4. 4

    Where you are

    Ask from a goal or your ledger and Lumi knows you mean “this goal” or “this month’s spending”.

AI you can trust

Grounded answers, undoable actions,
visible costs.

  • Every claim has a source

    When your data is involved, answers cite it — “from your journal on March 12”, “September food · 43 transactions” — one tap to the original.

  • Numbers are computed

    Amounts, durations, savings projections and statistics come from deterministic tools; the model only explains them. Expand key numbers to see the inputs.

  • Lumi proposes, you decide

    Single, undoable actions can run automatically with Undo if you choose; batch changes, anything affecting others and anything irreversible always ask first.

  • Activity log

    What Lumi read, did and created — all traceable.

  • Restrained initiative

    At most two proactive messages a day (adjustable 0–5), never in quiet hours, each with “why am I seeing this?”.

  • Honest

    When Lumi doesn’t know, it says so; when it’s unsure, it shows its assumptions. It never makes data up — and if no model is available, AI features simply switch off instead of showing simulated answers.

Models

The right model for each task,
routed by region.

Lumi connects to DeepSeek, Claude and OpenAI through its own model gateway. Auto mode picks the best value for each task; members can prefer a model among those available in their region.

Data regionModel providersNotes
China (in preparation) DeepSeek (primary), Qwen (backup) Once it launches: only domestically filed models, with data and calls staying in China and model names and filing numbers shown in the app. Until then, users in mainland China are served by DeepSeek from the Tokyo deployment, with a keyword content filter on AI inputs and outputs.
Global OpenAI, DeepSeek, and Anthropic Claude (once eligibility is confirmed) Available providers depend on your country or region.
Hong Kong & Macau DeepSeek and other providers that serve the region Anthropic and OpenAI do not currently support Hong Kong or Macau.
TaskTier
Capture parsing, journal tags Standard
Morning briefing, evening review Standard
Chat Standard, stepping up for complex analysis
Plan My Day, weekly insights Advanced
Growth plans, big decisions, yearly review Flagship (estimate first)

Coming later: on devices and in regions where Apple Intelligence is available, some light tasks will be able to run on Apple’s on-device and Private Cloud Compute models — without using your tokens.

Model names and routing are as of September 2026 and change with providers and evaluations; every change to models, prompts or routing must pass our evaluations first.

How tokens work

Privacy & safety

Your data,
with your consent first.

Explained before enabled
Before you first use cloud AI, Lumi explains what it will read, which provider receives it and why. AI stays off until you agree.
Per-module permission
Journal, docs, tasks, calendar, money, habits, goals and plans can each be switched on or off; health and email need separate permission.
Exclude any entry
Mark any journal entry or document “Don’t let Lumi read this”, and it is never retrieved or sent to a model.
Only what’s needed
Just the data a task needs; phone numbers, emails, ID and card numbers are masked before anything reaches a model.
Never used for training
Lumi never trains models on your data, and our provider agreements rule out training and keep retention minimal.
Memory under your control
Health conditions, religion, political views, sexual orientation and account numbers are never memorized automatically. Say “forget this” to delete.
Crisis care
If Lumi detects signs of a serious crisis, it stops everything else and gently shares local help lines — in the US, call or text 988.
Professional limits
Lumi is not a doctor, lawyer, financial adviser or therapist: no diagnoses, no specific investment picks, and a nudge toward professionals when it matters.
AI labels
Lumi’s answers carry the orb, and chat answers are labeled “AI-generated” by default. In a data export, AI-generated content is labeled “AI-generated” in Markdown and tagged ai_generated in JSON.
Outside content is just data
Text from web pages, emails and imported files is treated as data and can never trigger changes on its own.

AI features require you to be at least 18 (or the age required where you live).

Read the AI Service Terms

Questions about Lumi AI

  • What can Lumi do for me?

    Lumi is the AI companion that runs through the whole app. It plays four roles:

    • Capture assistant: turns a single sentence into a task, event, expense or journal entry
    • Daily secretary: morning briefings, evening reviews and help planning your day
    • Analyst: answers questions like “How much did I spend on food this month?” and shows its sources
    • Growth coach: breaks down life goals, builds growth plans and reviews your progress with you

    You decide which of your data it can read. Learn more about Lumi AI.

  • Which AI models does Lumi use?

    It depends on your data region:

    • Global region: OpenAI, DeepSeek, and Anthropic’s Claude once our eligibility is confirmed. Users in Hong Kong and Macau get DeepSeek and other providers that support those regions.
    • Mainland China region (being prepared): once it launches, only registered domestic models, with DeepSeek as primary and Alibaba Cloud Bailian (Qwen) as backup; data stays in China. Until then, users in mainland China are served by DeepSeek (requests processed in mainland China), and AI inputs and outputs pass through a keyword content filter.

    By default, Auto mode picks a suitable model for each task, and members can choose a preferred model. DeepSeek processes data in China. If no model is available, AI features pause; Lumi never shows simulated answers instead.

  • Can I use Lumi's AI in mainland China?

    Yes. Lumi’s AI features are available to users in mainland China: chat, briefings, planning and the other AI features are served by DeepSeek, a domestically filed large language model, with requests processed in mainland China and never sent to OpenAI or Anthropic. AI inputs and outputs pass through a keyword content filter, and requests or answers found to violate the rules are blocked or stopped.

    Lumi decides whether you’re in mainland China from the country or region and the time zone your device reports to Lumi. It doesn’t use GPS or your IP address.

    Lumi’s servers are in Tokyo, Japan, so we ask for your separate consent to the cross-border transfer before anything you create is uploaded. The generative AI service registration is in progress, and we’ll publish the registration details once it’s complete. See the AI Service Terms.

  • What is a token, and why do models use different amounts?

    Tokens are how Lumi measures AI usage. One standard token equals one token on our standard model, DeepSeek V4.1 Flash. Other models count with a multiplier based on cost, for example standard ×1, advanced ×4, premium ×8 and flagship ×20–40.

    You can expand any answer to see what it used, and big tasks show an estimate first so you can decide whether to go ahead. Multipliers are published, recalibrated when providers change their prices, and announced before they take effect.

  • What happens when I run out of tokens?

    Your life records aren’t affected. Tasks, calendar, journal, money, habits and every other non-AI feature keep working as usual.

    AI steps down gently: quick capture switches to local parsing, briefings become a plain summary of your data, and chat and analysis suggest a token pack or a membership upgrade. When your balance drops below 10%, Lumi reminds you once. Monthly tokens reset each billing period, and tokens you buy never expire.

  • Does Lumi train AI on my data?

    No. Lumi never uses your data to train any model, and our agreements with model providers say they won’t train on it either and will keep it for the shortest time possible.

    Before you use cloud AI for the first time, Lumi shows a single page explaining what data it reads, which provider receives it and why. AI stays off until you agree. See the AI Terms for details.

  • How do I control what Lumi can read?

    In Settings → Lumi, switch access on or off by module. Journal, docs, tasks, calendar, money, habits, and goals and plans are on by default once you agree to use AI. Health data, and the email features coming later, need separate permission.

    To keep a single journal entry or doc private, mark it “Don’t let Lumi read this.” Phone numbers, email addresses, ID numbers and card numbers are masked before anything is sent to a model.

  • What does Lumi remember about me, and can I manage it?

    Lumi picks up useful things from your chats, journal and planning interviews, such as your preferences, goals and the people in your life. Every memory shows its source, and you can view, edit or delete any of them in Lumi’s Memory. You can also just say “forget this” in a chat.

    Sensitive details such as health conditions, religion, political views, sexual orientation and financial account numbers are never memorized automatically unless you ask. The Lumi Activity Log shows everything Lumi has read, done and created.

  • Can Lumi change my data without asking?

    No. Lumi suggests; you decide.

    • Single actions you can undo, like adding a task or logging an expense: Lumi asks first. Only if you turn on auto-run for low-risk actions does it go ahead, with an Undo button.
    • Batch or structural changes: Lumi lists every item so you can check them and confirm.
    • Anything that affects other people or outside services, and anything irreversible such as deleting: Lumi always asks first.

    Deleted items go to Recently Deleted for 30 days.

  • Is Lumi always right? Can I rely on it for professional advice?

    Numbers about your own data, like how much you spent or how often you checked in, are calculated by the app, not guessed by the model. But AI-written text can still be wrong, so please double-check before any important decision.

    Lumi isn’t a doctor, lawyer, financial advisor or therapist. It doesn’t diagnose or prescribe, doesn’t recommend specific investments or when to buy or sell, and gives only general information and official sources on legal and immigration questions. If you’re going through a crisis, Lumi points you to local help lines.

  • Is there an age requirement for AI features?

    Yes. Lumi’s AI features require you to be at least 18, or the age required where you live.

    The minimum age to create an account is 14 in mainland China and 13 elsewhere, or higher where local law requires. If you’re under 18, you need a parent or guardian’s consent, and you can use only the non-AI features. Parents and guardians can reach us at support@xicoai.com.

  • Why do Lumi's answers carry an "AI-generated" label?

    So you can always tell what came from AI. In Lumi, AI answers carry the orb mark, and answers in conversations are labeled “AI-generated” by default.

    When you export your data, the AI-generated content in the archive, including Lumi’s answers in conversations, briefings and insights, AI summaries of journal entries and reviews, growth-plan analyses and generated plan text, and the memories Lumi created, is labeled “AI-generated” in the Markdown files and tagged ai_generated in the JSON files. The archive also includes metadata naming the service provider. We apply these labels under China’s rules on labeling AI-generated content.

Start today.
Make every week count.

Lumi for the web is open in early access. The iPhone, iPad and Mac apps are being polished and are coming soon to the App Store.