Post

Clawdbot Practical Guide: Building Your Dockerized Personal AI Command Center from Scratch

This tutorial will walk you through Clawdbot Docker deployment, Ollama model configuration, Slack integration, and custom Python Action development.

Published

In the previous deep dive, we learned about the power of Clawdbot (OpenClaw). Today, we’ll shed the theoretical coat and deploy this powerful AI assistant to your computer through real code and configuration.

Prerequisites: This tutorial requires basic command-line knowledge and Docker Desktop installed on your computer.

Phase 1: Environment Preparation and Basic Deployment

Clawdbot recommends Docker deployment to isolate environment dependencies from your host machine, keeping your system clean.

1.1 Hardware Requirements Check

While Clawdbot itself uses minimal resources, for smooth local large model operation (like Llama 3 8B), we recommend:

  • Memory (RAM): At least 16GB (24GB+ recommended for larger parameter models).
  • GPU: NVIDIA RTX 3060 or higher, or Mac M1/M2/M3 series chips (unified memory architecture is very LLM-friendly).
  • Storage: At least 50GB free SSD space (for model files).

1.2 Install Ollama (Large Model Backend)

Clawdbot is just the brain’s cortex; Ollama is the neurons. If your host is Mac or Linux, install Ollama directly:

# Mac / Linux
curl -fsSL https://ollama.com/install.sh | sh

If running Clawdbot in Docker, to allow the container to access the host’s Ollama, ensure Ollama listens on all interfaces, not just localhost.

  • Mac: Click Ollama icon in menu bar -> Quit. Then run in terminal: OLLAMA_HOST=0.0.0.0 ollama serve
  • Windows: Set environment variable OLLAMA_HOST to 0.0.0.0, then restart Ollama.

1.3 Start Clawdbot Container

Create a docker-compose.yml file for better configuration management:

version: '3.8'
services:
  clawdbot:
    image: openclaw/core:latest
    container_name: my-clawdbot
    restart: always
    ports:
      - "3000:3000"
    volumes:
      - ./clawdbot_data:/app/data  # Persistent data
      - /var/run/docker.sock:/var/run/docker.sock # Optional: manage other Docker
    environment:
      - OLLAMA_BASE_URL=http://host.docker.internal:11434 # Point to host Ollama
      - CLAW_API_KEY=your_secret_key_change_this

Run in terminal:

docker-compose up -d

Visit http://localhost:3000, you should see Clawdbot’s console interface.

Phase 2: Give It Intelligence (Model Configuration)

Enter console’s Settings -> Model Provider.

  1. Provider Selection: Choose Ollama.
  2. Model Pulling: Back to terminal, download the model first. 
    ollama pull llama3
ollama run llama3
  3. Configuration: In Clawdbot interface, fill Model Name as llama3.
  4. System Prompt (Persona): This is key. Enter:

    “You are Clawdbot, an efficient AI assistant running locally. Your characteristics are concise, precise, and highly executable. Please respond in English.”

Click Test Connection, if it returns “Success”, the brain connection is successful.

Phase 3: Connect to the World (IM Integration)

Using Slack as an example, as it’s most bot-friendly.

3.1 Create Slack App

  1. Visit api.slack.com/apps, click Create New App -> From scratch.
  2. On Socket Mode page, enable Socket Mode (no public IP needed).
  3. On OAuth & Permissions page, add these Scopes:
    • app_mentions:read (allow it to see @)
    • chat:write (allow it to speak)
    • files:write (allow it to send files)
  4. Install App to workspace, get Bot User OAuth Token (starts with xoxb-) and App-Level Token (starts with xapp-).

3.2 Configure Clawdbot

In Clawdbot’s Integrations -> Slack page:

  • Enabled: True
  • App Token: Fill in xapp-…
  • Bot Token: Fill in xoxb-…

Save and restart container. Now @Clawdbot in any Slack channel to see if it responds.

Phase 4: Advanced Play — Custom Python Actions

This is Clawdbot’s killer feature. Let’s write an Action: Query today’s GitHub Trending and push to Slack.

Create github_trending.py in ./clawdbot_data/actions directory:

import requests
from bs4 import BeautifulSoup
from claw_sdk import Action, Context

class GithubTrendingAction(Action):
    name = "get_github_trending"
    description = "Get today's GitHub trending repositories"
    
    def run(self, context: Context):
        url = "https://github.com/trending"
        response = requests.get(url)
        soup = BeautifulSoup(response.text, 'html.parser')
        cols = soup.select('article.Box-row')
        
        result_text = "*Today's GitHub Trending Projects*:\\n"
        for i, col in enumerate(cols[:5]): # Only top 5
            repo_name = col.select_one('h1 a').text.strip().replace('\\n', '').replace(' ', '')
            desc = col.select_one('p.col-9').text.strip() if col.select_one('p.col-9') else "No description"
            result_text += f"{i+1}. *{repo_name}*: {desc}\\n"
            
        return result_text

Restart Clawdbot. Now you can say: “What’s trending on GitHub today?”, it will call this Python script, scrape data, and send you formatted results.

Troubleshooting

  1. Can’t connect to Ollama: 99% network issue. Check if host.docker.internal works on your system (Linux users might need 172.17.0.1).
  2. Extremely slow response: Check CPU/GPU usage. If model is too large (like 70B) with insufficient VRAM, it switches to pure CPU mode, dropping to 0.5 token/s. Recommend switching back to Llama 3 8B or Phi-3.
  3. Missing Python dependencies: If your script uses requests or bs4, enter container to install: 
    docker exec -it my-clawdbot pip install requests beautifulsoup4

Conclusion

Congratulations! You now have a fully local, programmable, extensible AI hub. This isn’t just a tool installation process, but your first step from AI consumer to AI creator. Go explore, write more Actions, and make Clawdbot truly the commander of your digital life.