Learning / Workshop 2 — Singapore

Build Your Own Personal AI Assistant Bot

Build a private AI assistant bot you can message on Telegram — one that runs on its own always-on computer in the cloud, so it answers even when your laptop is closed. You'll set it up for one personal job, with clear rules for what it can do, what it must ask first, and where it should stop.

More hands-on than Workshop 1 — but you won't be typing terminal commands. Codex, an AI helper on your laptop, does the setup while you create the accounts and approve each step. Max 5 people per group in Singapore.

July sessions done · new dates coming in August

Next datesAugust 2026 — TBA
Time1:00–4:00pm
WhereClementi, SG
Length3 hours
SeatsMax 5
PriceS$199
Interested in future workshops →
Important — bring a personal laptop you can install software on. You'll install Codex before class, and locked-down work or corporate laptops usually block it.
Want a simpler option with no cloud setup? Workshop 1 covers reusable AI workflows — no bots, no VM, no tool connections. See Workshop 1 →
Looking to build something for your business instead? See AI solutions for small and medium businesses → — that's a tailored implementation, not a workshop.

From the June workshop

What people said

At the June workshop, every participant left with a working bot — and a Codex watchdog set up to keep an eye on it.

“I never imagined myself getting this ‘technical’ as a non-tech person. Oanh's AI workshop made it hands-on, can-do. Can't believe I have my own personal AI baby now.”
Xuan N. · Workshop 2 participant

The setup

What you'll build today

Six moving parts, one private bot. Here's how they fit together:

Telegram

Your front door — message the bot like a person. Optional: the OpenClaw mobile app as a second front door.

Google Cloud VM

A small always-on cloud computer you own — so the bot answers even when your laptop is closed.

OpenClaw

The bot engine. It runs on the VM and is the brain behind your assistant.

Codex

Runs on your laptop. It does the setup and the troubleshooting — you approve as it goes.

GitHub

Stores your bot's editable instruction files, so you can change its behaviour from your browser.

You

You hold the accounts, the secrets, and the final say. You set approval rules so risky actions wait for your decision.

Fit check

Is this for you?

This workshop is for you if

  • You've used AI tools and want a persistent personal assistant
  • You can create accounts for Telegram, Google Cloud, GitHub, and the class AI provider
  • You're happy to follow guided steps in Codex and approve what it does
  • You'll handle tokens and API keys carefully
  • You want the bot to help, but still ask before doing anything risky

Skip this workshop if

  • You'd rather a simpler option with no cloud setup — Workshop 1 covers reusable AI workflows instead.
  • Your only computer is a locked-down work or corporate laptop — you need one you can install Codex on.
  • You don't want to create a cloud VM or handle account setup.
  • You want a fully autonomous assistant that runs your life unattended.
  • You want a bot for your business — customer-facing or staff workflows need a tailored approach. See AI solutions for small and medium businesses instead.

Why this exists

Why this exists

Some AI tasks are fine inside Claude, ChatGPT, or Gemini. You paste in messy input, get a useful output, and move on. That's Workshop 1.

But some personal tasks keep coming back:

For those, it helps to have a bot you can message — one with its own context, rules, memory, and (eventually) light personal tool access.

I built bots for my own recurring personal admin. Workshop 2 is the practical version: not a magic fully autonomous assistant, but a small private bot that does one personal job well and safely.

How it works

Codex does the technical part

You won't memorise commands. You give Codex the right information — it runs the setup, reads the errors, and explains the next step. You stay in charge of secrets, accounts, and approvals.

You do

  • Create the accounts (Google Cloud, Telegram, GitHub)
  • Make the Telegram bot in BotFather
  • Paste the access key Codex gives you into Google Cloud
  • Hold the secrets — paste tokens only when asked
  • Approve anything that changes the system

Codex does

  • Set up secure login to your VM
  • Install OpenClaw and run onboarding
  • Read errors and explain them in plain English
  • Guide Telegram pairing
  • Suggest the smallest next fix — then wait

The flow

The setup map

First we make the bot reply. Then we make it useful.

  1. Create your Telegram bot token
  2. Create a Google Cloud VM
  3. Sign in to Codex on your laptop
  4. Codex creates an SSH key
  5. You paste the public key into Google Cloud
  6. Codex tests the connection
  7. Codex installs OpenClaw on the VM
  8. Connect Telegram (or pair the app)
  9. Start your GitHub instruction files
  10. Teach the bot one narrow job
  11. Set up a diagnose-only watchdog — later

Outcomes

What you'll leave with

What to expect (honestly)

This is a setup workshop. The win is one private bot, reliably alive on its own cloud computer, doing one personal job well — a real working thing, deliberately narrow.

  • You'll leave with: a private Telegram bot that replies to you, running on your own always-on Google Cloud VM, configured for one recurring personal job with clear rules.
  • You won't leave with: a fully autonomous assistant, or a hub of connected tools. Calendar, reminders, and file access are taught as your next layer — you might wire one up in class, but they're not the day's goal.
  • Setup is the real work. Most of it is Codex driving your VM while you watch and approve. Your laptop's OS barely matters — you mainly need a browser, Telegram, and Codex.

One bot, alive and trustworthy, beats five half-wired integrations. We get the foundation right.

Example projects

Example bot jobs

Good Workshop 2 projects are all personal:

Each of these starts as a bot you message. Pick a job that's useful even as chat-only — reminders and tool access come later.

Follow-through bot

You mention something you need to do; it clarifies the deadline and checks back later.

Appointment prep bot

You tell it about an upcoming appointment; it builds a checklist of questions and reminds you the day before.

Event / ticket reminder bot

You send it interesting events; it remembers deadlines, drafts calendar entries, and reminds you before tickets sell out.

Personal reading / research bot

You're learning about a topic over weeks; it keeps a running thread, suggests one next step at a time, and doesn't dump everything at once.

Household decision bot

You log open family questions and decisions; it remembers what's pending and surfaces it back at the right moment.

Packing / travel prep bot

You share a trip; it turns it into a packing list and surfaces reminders before you leave.

Fruit group-buy helper

You run a weekly neighbourhood group buy; it parses messy seller text into clean item/price rows, drafts summary messages, and flags unclear items — you approve everything before sending.

Curriculum

What we'll cover

Create your bot's front door

Make a private Telegram bot with BotFather, and learn why the token is a password.

Create the cloud computer

Set up a small Google Cloud VM with a standard workshop username — and a budget alert so there are no surprises.

Hand setup to Codex

Codex creates and tests secure SSH access, then installs OpenClaw on the VM. No gcloud, no manual terminal.

Verify the green checkpoints

SSH works, OpenClaw installed, the bot is running, model access understood, Telegram paired, bot replies.

Write the bot's instructions

Start AGENTS.md and USER.md in a GitHub repo Codex copies to the VM — the job, tone, boundaries, and failure cases.

Teach one narrow personal job

Pick a recurring task. Write context, rules, boundaries. Test the happy path AND the "ask before acting" cases.

Safety and permission levels

The green / yellow / red model: read & draft freely, write with care, never send / delete / spend without approval.

Debugging and the watchdog

Codex is your first responder for setup errors — and later, a diagnose-only watchdog that checks the bot and waits for your OK.

The business boundary

Why business assistants need a different setup, and where to go if you want one.

Safety

Permission levels

Your bot earns trust before it gets power. Every action falls into one of three buckets:

Green

Do freely

  • Read and draft
  • Answer in chat
  • Summarise and explain

Yellow

Ask first

  • Send a message
  • Schedule something
  • Edit a file

Red

Never without you

  • Make a purchase
  • Delete data
  • Add new access

Your instructor

About your instructor

Oanh founded Paper & Pixel, a Singapore applied AI studio making AI useful, understandable, and safe for real life and small business. Its work spans practical products such as StencilCV and Zesca, hands-on learning, and tailored solutions for small and medium businesses.

Her background includes Google, McKinsey, and Intuit. In the workshop, the emphasis is practical: build one narrow bot, test it, set clear approval boundaries, and keep people in control.

Logistics

Logistics

August 2026 — S$199

Date / timeAugust 2026 — dates to be announced, 1:00–4:00pm SGT
LocationClementi, Singapore — exact venue confirmed by email
FormatLive in-person workshop
Duration3 hours
Capacity5 seats (intentionally small)
PriceS$199 per person
RefundsRefundable when requested at least 7 days before; transferable to another attendee before the session

The July session has wrapped. New August dates are coming soon — join the interest list and you'll be first to know when booking opens.

Interested in future workshops →

What to bring:

You'll set your bot up with your own AI provider from the start — so it keeps working after class, with no temporary keys to switch over later.

FAQ

FAQ

How long is the workshop?

3 hours of focused, hands-on work. The first stretch is setup — creating your Google Cloud VM and letting Codex install OpenClaw on it. The rest is teaching the bot one job, writing its rules, and testing failure cases. Bring a fully charged laptop.

Will I leave with a working bot?

Yes — a private Telegram bot that replies to you, running on your own Google Cloud VM, with one configured personal job. See "What to expect" above for scope detail.

Do I install OpenClaw on my laptop?

No. OpenClaw runs on your Google Cloud VM. Codex runs from your laptop and does the install and troubleshooting for you.

Do I need to know terminal commands or how to code?

No professional coding background, and you won't memorise commands. You follow Codex-guided steps, paste specific values, handle logins and tokens, and read errors with help.

Do I need a Google Cloud account? Will it cost money?

Yes. A free trial may cover the workshop setup, but cloud charges depend on your account and usage. We set a budget alert in class and show you where to review costs before you leave.

What's Codex, and do I need it installed?

Codex is the assistant on your laptop that creates secure access to your VM, installs OpenClaw, checks status, and helps debug — and later watches the bot. Bring a laptop with Codex installed and signed in.

Does my bot run all the time?

Yes — it runs on your always-on Google Cloud VM, so it answers even when your laptop is closed, as long as the VM is up and your own provider key remains active. You configure that key during the workshop. The bot is still not fully autonomous — its rules require approval before risky actions.

Which AI model and bot runtime will we use?

We use OpenClaw on your VM, and Codex on your laptop to drive the setup. You connect your own AI provider — OpenAI is the simplest since you'll already use it for Codex, though Anthropic, Google, or OpenRouter also work. You set this up from the start, so there's nothing temporary to swap out later.

Will my bot keep working after the workshop?

Yes. You set it up with your own AI provider account from the start, so there's nothing to swap out afterwards. As long as your VM is running and your provider account is active, your bot keeps replying.

How should I handle my API keys, tokens, and SSH keys?

Treat your Telegram bot token, AI API key, and SSH private key like passwords — never post them in chats, screenshots, shared docs, or GitHub. Your public SSH key is the one safe thing to paste into Google Cloud. We cover safe storage in class.

What does "private bot" actually mean?

The Telegram bot and OpenClaw workspace run on your own VM and aren't meant for public/group use. Your messages may still be sent to whichever AI model provider you use — Anthropic, OpenAI, Google, or OpenRouter — depending on your key and settings.

Refund policy?

Refundable when requested at least 7 calendar days before the workshop. Before the session, you may also transfer your place to someone else. See the Returns & Refund Policy.

Can I use this for my business?

Workshop 2 is for personal bots only. For customer-facing or staff workflows, see AI solutions for small and medium businesses.

After the workshop

After Workshop 2

Most people use Workshop 2's bot for one personal recurring job — follow-throughs, appointment prep, weekly reading threads.

If you run a small business and want help applying the pattern to a real workflow, that's a different kind of project. Business solutions need workflow mapping, the business's actual context and tools, appropriate permissions, and human approval for decisions involving customers, money, official records, or reputation. Workshop 2 doesn't cover that scope.

For business builds, see AI solutions for small and medium businesses →.

For small and medium businesses Running a business? Explore a tailored AI solution built around your actual workflow, rules, and tools.
Explore solutions →

Ready to build your bot?

The July session has wrapped. New dates are coming in August — join the interest list and you'll be first to know when booking opens.

Interested in future workshops →

Questions? Email