Software 3.0 deployed locally in Hong Kong.
The gentler way to work with AI.
Personal computers were meant to be personal. Mona takes that seriously.She is the AI that remembers what private really means.
The Software 3.0 era
Plain English is the new programming language.
Software 1.0
Code
Explicit rules in syntax a human must learn.
Software 2.0
Data
Learned weights. The program becomes invisible.
Software 3.0
Schedule my Monday standup, then send the agenda to the team.
Prompts
Plain language. Talk, and the model acts.
Mona is Software 3.0. You program her by talking.
AI researcher Andrej Karpathy frames the last fifty years of computing as three eras. Mona belongs to the third — and so do you, the moment you start talking to her.
›What changes in Software 3.0
Pre-trained language models become general-purpose computers, programmable in English. The system prompt is the program; the context window is the runtime. Non-engineers ship working software by writing clearly.
A Day with Mona
Mona holds the loose ends, so you can keep your head clear.
Morning Briefing
Mona reviews new mail and pending approvals, plus the reminders and follow-ups you've asked her to track (including scheduled check-ins). A structured summary arrives on Telegram before your first coffee.
3 urgent messages (2 Telegram, 1 email)
2 pending approvals from yesterday
Reminder: meeting with Leung & Partners at 2:00 PM
Would you like me to draft replies to the urgent items?
Inbox Triage
Three client emails land while you are in a call. Mona sorts them by urgency, drafts replies, and waits for your approval before anything is sent.
High priority: Confirm viewing at 3 PM Thursday
Draft ready: Send floor plan PDF
Draft ready: Acknowledge deposit timeline
Nothing has been sent yet.
Meeting Prep
Before your 2 PM meeting—tracked on the reminder list you keep with Mona—she surfaces the contact's history, prior decisions, and relevant documents from memory.
Last contact: 12 March — discussed lease renewal terms
Open items: tenant improvement budget (HK$180K)
Relevant doc: Q1 property report (shared 8 March)
No outstanding commitments from your side.
Commitment Follow-Up
Yesterday you promised a colleague you would send the updated proposal. Mona notices it is still pending.
Status: not yet sent.
Shall I draft it now, or set a new deadline?
End-of-day batch (custom skill)
You and Mona shape how you like to close the day; she can save that rhythm as a reusable skill. Running it batches what piled up while you were heads-down and parks non-urgent follow-ups for the morning.
- Newsletter from HKEX (read later)
- Vendor follow-up from Wong Kee Supplies
- Email inquiry — product question
- Reminder: renew building access card by Friday
Parked for tomorrow morning unless you want one handled now.
Weekly Report
At the end of the week, Mona compiles a narrative summary of what you accomplished, what is still pending, and opportunities she spotted.
Completed: 14 tasks, 23 messages sent, 2 viewings
Pending: proposal for Sarah Wong, Tsuen Wan deposit follow-up
Opportunity: Wong Kee Supplies offered 15% seasonal discount — expires 5 April
Full report saved to memory.
You stay in control.
Mona handles the busywork, but every decision is still yours. Your conversations stay on your Mac, sensitive accounts remain private, and anything risky waits for your approval first.
MonoClaw is provisioned with a local Gemma model through LM Studio that you can use free, forever. Add hosted frontier reasoning only when the task calls for it; MonoClaw adds no model usage surcharge.
Work stays on your Mac
Everyday tasks run through a local AI model included with your MonoClaw. Nothing leaves your desk unless you explicitly choose a cloud provider.
Your passwords stay hidden
MonoClaw never sees your login details in plain text. Account credentials are injected only where needed, and never fed into the AI as readable prompts.
You approve anything sensitive
Before MonoClaw sends a message, books an appointment, or runs a system task, it asks you first. You can approve, edit, or stop any action at any time.
Built-in encryption by default
Your setup keys and account secrets are stored with strong encryption and restricted access. Even if something goes wrong, error messages are scrubbed clean of sensitive data.
Edge-first by design
Your computer can run this. Not one litre lost to cooling.
Most AI assistants you have used are streamed to you from a hyperscale data centre. Every prompt travels the public internet, gets queued behind tens of thousands of other prompts, and answers come back via servers cooled by water that does not come back.
MonoClaw runs the other way. The Apple Silicon you already own is, by 2025 standards, a serious inference computer. Mona's day-to-day work — triage, memory, drafting, approvals — happens on your Mac, at your desk, without leaving the room. You own the silicon. You own the data. You own the model running on it.
Mac mini M4 peak power1
65 W max
A small desk lamp's worth of power, even under load.
Global data centre electricity2
460 TWh in 2024, on track for >1,000 TWh by 2026
Roughly Japan's entire annual electricity consumption — driven largely by AI.
U.S. data centre water3
17.4 billion gallons in 2023
Up to 85% evaporates for cooling and never returns to the watershed.
Your bandwidth is fixed. Your reach doesn't have to be.
Start where you are. Scale into an operator.
Catch what slips. Prep without panic.
Clutter → Clarity
Keep every thread moving without living in your inbox.
Scattered → Flowing
See what's due. Skip the status theater.
Overloaded → In Control
How It Works
From purchase to a quality-checked Mona setup in four steps.
Choose the software package with a one-time fee of HK$28,888.
Buy a compatible Mac mini or iMac directly from Apple. Mac mini starts at HK$6,099.
Ship the device to us. We install Mona, configure the system, and prepare it for your workflow.
We run comprehensive checks, then deliver the setup back ready for use.
Mona is ready when you are.
MonoClaw software from HK$28,888. Runs on your Mac. No MonoClaw subscription.