Building OpenClaw: From Zero to One

I walked newcomers through getting their first OpenClaw assistant running and connected to Telegram, from the setup choices to the first conversation.

SQ Collective · Singapore · 27 February 2026

Speaker discussing security beside presentation screens with an audience in the foreground
Building OpenClaw: From Zero to One presentation, Singapore, 27 February 2026.

From Setup to a First Conversation

At the first OpenClaw Singapore meetup, I explained how to choose where the assistant runs and which AI service it uses, considering cost, capability, and the work of maintaining it. The walkthrough then put an assistant on a cloud server and connected it to a Telegram bot so a user could start asking it to do things.

I also introduced instructions that shape the assistant's behavior and the difference between a task scheduled for a fixed time and a regular check for updates. Access controls, trusted tools, and keeping the installation updated were part of the setup, not steps to leave until after connecting the assistant.

Bringing a First Assistant Online

Hosting the Gateway and Choosing a Model

I compared virtual machines, containers, and managed services by setup and maintenance effort. The walkthrough used Alibaba Cloud's Simple Application Server with an OpenClaw image, covering ports, gateway access, and model configuration with Qwen 3.5 Plus as the default. Hosting determines where the assistant runs; model selection separately weighs context windows, price, performance, and token efficiency.

Connecting and Pairing Telegram

BotFather supplied the Telegram bot and token. Approving the pairing code in the terminal connected the user to the assistant, rather than treating possession of the bot's chat address as sufficient access. Domains, exposed ports, trusted skills, and updates were part of the setup discussion.

Behavior and Recurring Work

Soul instructions shaped how the assistant responded, while cron and heartbeat determined when work began. The cron example parsed a spreadsheet at 9 AM; a heartbeat supported periodic checks and proactive updates. Separating behavior from scheduling made it possible to change the assistant's instructions without confusing a fixed appointment with a recurring check.

Return to Speaking