A complete hands-on walkthrough from server choice and VPS security to OpenClaw installation, Anthropic setup, Telegram pairing, troubleshooting and the final security audit.
This is the practical follow-up to the OpenClaw explainer. Instead of only talking about what a self-hosted assistant could be, the video walks through the full setup from a fresh server to a working Telegram-connected assistant.
The most important decision comes first: where should OpenClaw run? Your main laptop is convenient, but risky because an agent with system access can reach too much. A dedicated local device gives more control, but needs more network setup. A VPS is the cleanest tutorial path because it is isolated, reachable and reproducible.
The video then builds a basic security baseline before installing anything important: SSH keys, updates, firewall rules, auto-updates and password-login hardening. Only after that does OpenClaw get installed and configured.
The assistant is connected to Anthropic for model access and Telegram for the chat interface. The setup includes common failure points, how to read OpenClaw diagnostics, how to restart the gateway and why the built-in security audit belongs at the end of the setup rather than later.
| Area | What the video covers | Why it matters |
|---|---|---|
| Host | VPS, local device or main machine | Controls risk, availability and how isolated the assistant is. |
| Server | Fresh Hetzner VPS with SSH access | Gives the tutorial a reproducible baseline. |
| Security | Firewall, updates, SSH keys, password login disabled | Reduces obvious exposure before the agent gets tool access. |
| OpenClaw | Installer, onboarding, gateway status and diagnostics | Turns the server into a working assistant environment. |
| Model | Anthropic API key and model choice | Keeps testing cheap while still allowing stronger models for real work. |
| Interface | Telegram bot creation and pairing | Lets you chat with the assistant from anywhere. |
| Audit | OpenClaw security audit and fixes | Checks for dangerous defaults before daily use. |
The video compares three options: your main computer, a dedicated local device such as a Raspberry Pi or Mac Mini, and a VPS. The pragmatic recommendation is to avoid your main machine, especially when the agent has broad system access.
No, but the walkthrough uses a VPS because it is reachable, isolated and easier to keep running. A dedicated local device can also work if you know how to expose or access it safely.
The tutorial recommends using a cheaper model for testing and Sonnet for real work. It also recommends setting a strict spending limit on the Anthropic account before connecting the API key.
OpenClaw can access tools and files, so configuration mistakes matter. The security audit checks for common problems after setup and should be run immediately before treating the assistant as usable.
Summary of all thirteen chapters. Timestamps jump straight to the matching point in the video. This video currently only has auto-generated subtitles, so there is no verbatim transcript here yet.
The opening sets the goal: a private AI assistant running on your own server and reachable through chat. It connects this tutorial to the previous OpenClaw explainer and makes clear that the result should be working, secure and under your control.
The video compares the main machine, a dedicated local device and a VPS. The key point is isolation: OpenClaw can read files, run commands and access a system, so running it on the computer that holds all your work is the most dangerous option.
The secure connection starts with SSH keys rather than passwords. The walkthrough shows how to find an existing public key or create a dedicated one for this server, then use it to connect without exposing password login as the main access method.
The tutorial creates a small VPS and walks through the basic choices: location, image, key, name and initial access. The point is not Hetzner specifically, but a clean server you can reproduce and isolate from your personal computer.
Before installing OpenClaw, the server gets updated and hardened. The video covers unattended security updates, firewall defaults, allowing SSH and reloading the SSH service after disabling password login.
OpenClaw is installed with the official installer. The walkthrough shows the one-command installation, then moves into onboarding and the daemon setup so the gateway can keep running beyond the current shell session.
The configuration step wires the assistant together: gateway status, initial settings and the practical checks that confirm the service is reachable and ready for a chat interface.
The video shows where to get an Anthropic API key and why a spending limit matters before testing an agent. The pragmatic split is simple: use a cheap model to verify the setup, then a stronger model when the workflow deserves it.
The assistant becomes usable through Telegram. The walkthrough creates a bot, connects it to OpenClaw, lists the pending pairing and approves the pairing code so messages reach the gateway.
The setup does not assume everything works on the first try. It covers diagnostic commands, automatic fixes, gateway restarts and logs, which are the fastest way to separate a model issue from a server or gateway issue.
The demo proves that the assistant is not only connected to a chat app. It can access files in the environment it runs in, which is exactly why the earlier security and isolation decisions matter.
The security audit is the last required checkpoint. It finds configuration problems and gives you a chance to fix critical warnings immediately, before treating the assistant as something you can rely on.
The closing points toward skills, automation and the daily workflows that make a self-hosted assistant useful. The setup is the baseline, but the real value comes from connecting it to the parts of your work where context and access matter.
I work 1:1 with freelancers, consultants, coaches and small teams on practical AI workflows, including the question of what belongs on your own infrastructure and what belongs in the cloud.
Book a free intro call