TUTORIAL 04 · AI / AGENTS · Build time: about an hour · Running cost: your model API usage · Level: comfortable in a terminal
What you'll build
A personal AI assistant that runs on your own computer (a spare Mac mini is ideal), which you message from Telegram like a colleague. It can read and write files, call APIs and run commands, and it can run jobs on a schedule. The first job you'll give it: a 7am briefing with your calendar, your due tasks and the news you care about.
We'll use OpenClaw, the open-source project that went viral in January as Clawdbot (briefly renamed Moltbot). The same pattern works with any local agent framework.
Why it matters
Subscriptions turn into scripts. Simple Zapier-style jobs (watch a feed, summarise, send) become scheduled tasks on hardware you already own.
Your data stays local. Memory is plain files on your disk, not a vendor's database.
It builds its own tools. Ask for a new capability and the agent can write the integration itself, which is both the magic and the risk.
The honest trade-off. An agent with shell access and your API keys is a high-value target. Security researchers have already found info-stealer malware hunting for these installs, and malicious community "skills". This tutorial puts the lock-down steps first, not last.
What you need
A computer that stays on: ideally a separate Mac mini or Linux box, not your work laptop.
Node.js 22 or later.
An API key for a capable model (Anthropic, OpenAI or another supported provider).
A Telegram account.
Subscribe to our premium content to read the rest.
Become a paying subscriber to get access to this post and other subscriber-only content.
Upgrade