How to Set Up Your Own AI Personal Assistant

A production-grade guide to running your own AI assistant on a $7/month VPS using OpenClaw and Claude.

1. Infrastructure: Running Your Own Always-On AI Layer

We finally have the ingredients to run a real personal AI assistant. Frontier models like Claude Opus 4.6 can reason across long context, synthesize complex information, and generate structured outputs. Orchestration layers like OpenClaw abstract the plumbing. Cheap cloud compute makes it practical to keep everything online 24/7.

The result is simple but powerful: an assistant you can message from your phone that helps you think faster and execute with less friction. Research, analysis, drafting, planning, coding support, decision structuring --- all through a single interface that is always available.

The infrastructure to run this is not expensive. In my case, I use a Hetzner CAX21 instance, which costs €6.99 per month (roughly $7--8 depending on exchange rates). The machine runs on ARM architecture and includes 4 vCPU, 8GB RAM, and 80GB of NVMe storage. This is more than enough because no heavy AI inference happens locally. The server runs coordination logic only. The actual reasoning happens through the Claude API.

In practice, the VPS handles session state, Telegram updates, and model orchestration. CPU usage remains modest because the workload is mostly network-bound rather than compute-bound. You are not running a model cluster. You are running coordination infrastructure.

That distinction is what keeps the cost low and the architecture elegant.

2. Deploying the VPS and Preparing the Environment

Go to:

https://console.hetzner.cloud/

Create a new server with:

Ubuntu 22.04 LTS
Instance type: CAX21
Add your SSH key (recommended)

If you add an SSH key during provisioning, you will be able to log in without a password. Otherwise, Hetzner will generate a root password.

Connect via SSH:

ssh root@YOUR_SERVER_IP

If using password authentication, you will be prompted to paste the root password.

Once inside the machine, update and prepare the environment:

apt update && apt upgrade -y
apt install -y git curl ca-certificates

At this stage, you have a clean Linux environment running continuously in the cloud.

3. Installing Node.js and OpenClaw

OpenClaw runs on Node.js. Install Node 20 LTS:

curl -fsSL https://deb.nodesource.com/setup_20.x | bash -
apt install -y nodejs
node -v
npm -v

Because the CAX21 runs on ARM64, if you choose Node 22 instead of 20, you may need to install build tools to compile native dependencies:

apt install -y build-essential make g++ pkg-config python3 libopus-dev

Now install OpenClaw globally:

npm install -g openclaw

If installation throws errors, copy the full error output and paste it into Claude. Frontier models are extremely good at debugging Node dependency issues, ARM build failures, missing packages, and PATH problems. Most installation issues can be resolved this way quickly.

Verify installation:

openclaw --help

4. Configuring OpenClaw and Connecting Claude Opus

Instead of manually exporting environment variables, use the OpenClaw configuration wizard:

openclaw config

Follow the prompts carefully.

When asked for your model provider credentials, enter your Anthropic API key.
When prompted to select a model, choose Claude Opus 4.6.

The configuration will be stored under:

~/.openclaw/openclaw.json

At this point, your assistant has a reasoning layer connected.

5. Creating and Connecting Your Telegram Bot

Open Telegram and search for @BotFather.

Start a conversation and run:

/newbot

BotFather will ask for a display name and a username (which must end in "bot"). Once created, it will generate a token. Copy this token.

Back on your VPS:

openclaw channels login --channel telegram

Paste the token when prompted.

Choose pairing mode for DM policy.

Now send a message to your bot from your Telegram account. You will receive a pairing code. Approve it:

openclaw pairing approve telegram YOUR_PAIRING_CODE

If you want to restrict usage strictly to your own Telegram ID:

openclaw config set channels.telegram.dmPolicy allowlist
openclaw config add channels.telegram.allowFrom "YOUR_TELEGRAM_USER_ID"

6. Starting the Gateway and Keeping It Online

Start the OpenClaw gateway:

openclaw gateway --host 127.0.0.1

To keep it running even after closing SSH, use tmux:

apt install -y tmux
tmux new -s assistant
openclaw gateway --host 127.0.0.1

Detach with:

CTRL + B
then press D

Reattach later:

tmux attach -t assistant

7. What Happens When You Send a Message

When you message your Telegram bot, the request flows to the OpenClaw gateway on your VPS. The gateway attaches session context and forwards structured input to Claude Opus. Claude performs inference remotely and returns a response. The gateway routes that response back through Telegram.

The server remains lightweight because it performs coordination, not inference. Claude handles reasoning at scale via API. The VPS simply orchestrates the interaction.

For roughly seven dollars per month in infrastructure, you now operate a persistent AI layer that is always reachable, isolated from your laptop, and fully under your control.

← Back to all posts