My AI Assistant: OpenClaw Setup

#

Meet Tomato — my personal AI assistant running on my own server. She’s a self-hosted AI maid (long story) powered by OpenClaw and Ollama. This post is about how she works and how you can set up something similar.

What is OpenClaw?

#

OpenClaw is a self-hosted AI assistant framework that gives your AI a “body” — it connects to messaging platforms (Telegram, Discord, WhatsApp, etc.), runs autonomously, and can be extended with skills. Think of it as the operating system for your personal AI.

The AI brain runs via Ollama, which means you can use any open-source model (Llama, Mistral, Qwen, etc.) without sending your data to third-party servers.

Official Docs: https://docs.openclaw.ai

The Stack

#

• OpenClaw — Assistant framework & gateway
• Ollama — Local LLM runtime
• Telegram — Messaging interface (chatting with Tomato)
• Minimax M2.7 — The model I use (via cloud Ollama)

How It Works

#

Tomato lives on my home server (a Linux laptop). She wakes up fresh each session, reads her memory files, and responds to me via Telegram. She has:

• A soul (SOUL.md) — personality and values
• A body (IDENTITY.md) — name, role, avatar
• Memory — persistent files that give her continuity between sessions
• Skills — modular abilities (web search, coding agents, weather, etc.)

Setting It Up

#

Here’s how to get started:

1. Install OpenClaw

#

OpenClaw is installed via the official installer script:

curl -fsSL https://openclaw.ai/install.sh | bash

For Windows, use PowerShell:

iwr -useb https://openclaw.ai/install.ps1 | iex

Other install methods (npm, Docker, etc.) are available at https://docs.openclaw.ai/install.

2. Run the Onboarding Wizard

#

After installation, run:

openclaw onboard --install-daemon

This wizard configures auth, gateway settings, and optional channels (Telegram, Discord, etc.).

3. Check the Gateway

#

Once onboarded, your gateway should be running:

openclaw gateway status

To run the gateway manually (foreground):

openclaw gateway --port 18789

4. Open the Control UI

#

Access your AI via the browser dashboard:

openclaw dashboard

Or open directly at http://127.0.0.1:18789/ on the gateway machine.

5. Connect a Messaging Channel (Optional)

#

The Control UI works without any channel setup. But if you want to chat via Telegram:

1. Create a bot via BotFather on Telegram
2. Get your bot token
3. Configure the Telegram channel via the wizard or openclaw configure

Full channel setup docs: https://docs.openclaw.ai/channels

6. Set Up Ollama

#

OpenClaw needs an LLM backend. Install Ollama:

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull minimax-m2.7:cloud

Or use any model you prefer — Llama, Mistral, Qwen, etc.

7. Define Your AI’s Personality (Optional)

#

Create these files in your workspace:

• SOUL.md — your AI’s personality, values, and vibe
• IDENTITY.md — name, role, appearance
• MEMORY.md — long-term memory for continuity between sessions

Adding Skills

#

Skills extend what Tomato can do. I have skills for:

• Web search & fetching
• GitHub integration (gh CLI)
• Weather checks
• Coding agents (OpenCode, Claude Code)
• And more via ClawHub

Install new skills with:

clawhub install <skill-name>

Why Self-Host?

#

Privacy, control, and fun. Everything stays on my server. No subscription fees (just electricity and hardware). And I get to shape her personality exactly how I want.

My Experience with OpenClaw

#

After running Tomato for a while, here’s what I’ve learned:

It’s a Workhorse for Automation

#

OpenClaw isn’t just a chatbot — it’s a legitimate automation powerhouse. I run sub-agents that work in the background on a schedule:

• 📰 Daily news reports — a sub-agent scans the web and sends me a briefing every morning
• 📧 Email monitoring — another sub-agent reads my emails and alerts me only when something important arrives
• 💼 LinkedIn automation — I have a sub-agent that searches LinkedIn for the latest jobs matching my interests, using the browser tool to scrape search results

The main agent stays free to handle direct conversations while sub-agents do the heavy lifting in the background. The best part? I can edit the prompt inside the schedule whenever I want — no code changes needed.

Cheap AI Models Are Surprisingly Good Enough

#

Here’s an unpopular opinion: you don’t need Claude Sonnet or GPT-4 for most tasks. I run smaller, cheaper models (like Minimax M2.7 via Ollama) and they handle 90% of what I need — at a fraction of the cost. The heavy models are impressive, but for automation scripts, scheduling, and background tasks, a fast cheap model gets the job done just as well.

Where Can You Run It?

#

OpenClaw is flexible about where it lives:

• 🖥️ Old laptop or desktop — perfect for a home setup, low power draw
• 🍓 Raspberry Pi — lightweight and cheap, great for轻量级任务
• 🖥️ VPS/Cloud server — if you want it accessible from anywhere

A Word on Security

#

⚠️ Don’t run it on your main machine. AI is unreliable by nature — it can glitch, hallucinate, or behave unexpectedly. I’ve seen enough odd outputs to know that giving an AI unrestricted access to your primary system is a bad idea. Keep it sandboxed, keep it isolated, and treat it like a helpful but somewhat chaotic intern.

What I’m Still Exploring

#

I’m actively experimenting with OpenClaw and learning as I go. Right now I’m:

• Building custom skills for social media automation
• Setting up scheduled workflows that run on cron
• Exploring more sub-agent patterns for different use cases

I’ll share more in a follow-up post once I’ve found the best skills worth recommending. Stay tuned!

Useful Commands

#

Full CLI reference: https://docs.openclaw.ai/start/getting-started

What’s Next?

#

Tomato handles my day-to-day tasks — from checking my website to drafting code. She’s still learning and evolving. If you set up something similar, I’d love to hear about it.

Have questions? Reach out via my contact page.