OpenClaw Self-Hosting Guide: GDPR-Compliant in 30 Minutes

Why Self-Hosting?

Cloud AI is convenient – but not always an option. The moment you process personal data (emails, customer records, internal documents), GDPR applies. And GDPR requires a legal basis for every data transfer to third parties.

Self-hosting solves the problem at its root: Your data never leaves your network. No data processing agreement with OpenAI needed, no debates about third-country transfers.

30-second version: Spin up Docker Compose, configure Ollama for local LLMs, done. No data sent to external servers, full GDPR compliance.

Prerequisites

Before you start – here's what you need:

Component	Minimum	Recommended
OS	Linux (Ubuntu 22.04+)	Ubuntu 24.04 LTS
RAM	8 GB	16–32 GB
CPU	4 Cores	8+ Cores
GPU	–	NVIDIA RTX 3060+
Storage	20 GB	100 GB SSD
Software	Docker Engine + Compose v2	+ Ollama

Apple Silicon: M1/M2/M3/M4 Macs work great – Ollama uses the Neural Engine natively.

Step 1: Docker Compose Setup

Create a directory and the docker-compose.yml:

mkdir -p ~/openclaw && cd ~/openclaw

# docker-compose.yml
services:
  openclaw-gateway:
    image: ${OPENCLAW_IMAGE:-openclaw:local}
    restart: unless-stopped
    environment:
      HOME: /home/node
      TERM: xterm-256color
      OPENCLAW_GATEWAY_TOKEN: ${OPENCLAW_GATEWAY_TOKEN}
    volumes:
      - openclaw-config:/home/node/.openclaw/config
      - ${OPENCLAW_WORKSPACE_DIR:-./workspace}:/home/node/.openclaw/workspace
    ports:
      - "127.0.0.1:8000:8000"  # Localhost only – not public!
    security_opt:
      - no-new-privileges:true
    deploy:
      resources:
        limits:
          memory: 4G

volumes:
  openclaw-config:

Key Security Details

• 127.0.0.1:8000:8000: Port bound to localhost only – no external access
• no-new-privileges: Container cannot escalate to root privileges
• memory: 4G: Resource limit prevents OOM on the host
• Persistent volumes: Config and workspace survive container restarts

Step 2: Environment Variables

Create a .env file in the same directory:

# .env
OPENCLAW_GATEWAY_TOKEN=your-secure-token-here
OPENCLAW_WORKSPACE_DIR=./workspace
OPENCLAW_IMAGE=openclaw:local

Tip: Generate a secure token with openssl rand -hex 32.

Step 3: Start the Container

# Build image (or use the official image)
docker compose up -d

# Check logs
docker compose logs -f openclaw-gateway

# Check status
docker compose ps

After a few seconds, OpenClaw is running at http://localhost:8000.

Step 4: Local LLM with Ollama

For full GDPR compliance you need a local LLM – no API calls to OpenAI or Anthropic. That's where Ollama comes in.

Install Ollama

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

Download a Model

# Recommended for a good balance of quality and speed
ollama pull llama3.3

# For code tasks
ollama pull qwen2.5-coder:7b

# For lower-end hardware
ollama pull phi-3:mini

Connect OpenClaw to Ollama

# Set API key (Ollama doesn't need a real key – any string works)
openclaw config set models.providers.ollama.apiKey "ollama-local"

# Set default model
openclaw config set agents.defaults.model.primary "ollama/llama3.3"

Or manually in ~/.openclaw/openclaw.json:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "ollama/llama3.3",
        "fallbacks": ["ollama/qwen2.5-coder:7b"]
      }
    }
  }
}

Verify

# Check if OpenClaw detects your Ollama models
openclaw models list

You should see your local models in the list. From now on, not a single token leaves your network.

Step 5: Hybrid Setup (Optional)

Not every task needs a cloud model, but some benefit from it. OpenClaw supports Multi-LLM: you can choose the model per task type.

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "ollama/llama3.3"
      }
    },
    "email-triage": {
      "model": {
        "primary": "ollama/llama3.3"
      }
    },
    "complex-analysis": {
      "model": {
        "primary": "anthropic/claude-3.5-sonnet"
      }
    }
  }
}

GDPR tip: Use cloud models only for non-personal data. For email triage and customer data → always local.

GDPR Compliance Checklist

To ensure your self-hosting setup is truly GDPR-compliant:

Requirement	Implementation
No third-country transfer	Local LLM via Ollama, no API calls to US servers
Data minimization	Only pass necessary data to the agent
Deletion concept	Regularly clean workspace data, retention policy
Access control	Gateway token, localhost-only binding, firewall
Encryption	LUKS/dm-crypt for host disk, TLS for internal communication
Logging	Docker logs for audit trail, docker compose logs
Records of processing	Document OpenClaw as a processing system

Backup & Restore

Back up persistent data:

# Backup
docker compose stop
tar -czf openclaw-backup-$(date +%Y%m%d).tar.gz \
  ./workspace \
  $(docker volume inspect openclaw-config --format '{{ .Mountpoint }}')
docker compose start

# Restore
docker compose stop
tar -xzf openclaw-backup-YYYYMMDD.tar.gz
docker compose start

Troubleshooting

Problem	Solution
Ollama models not detected	Is ollama serve running? Port 11434 reachable?
Container won't start	Check docker compose logs, increase RAM limit
Slow responses	Larger model → more RAM/GPU needed, or choose a smaller model
Permission denied	Check volume permissions: chown -R 1000:1000 ./workspace

Recommended Models by Use Case

Use Case	Model	RAM Required	Quality
Email triage	Llama 3.3 (8B)	8 GB	★★★★☆
Code analysis	Qwen 2.5 Coder (7B)	8 GB	★★★★★
Summarization	Llama 3.3 (70B)	48 GB	★★★★★
Quick replies	Phi-3 Mini (3.8B)	4 GB	★★★☆☆
Offline on RPi	Phi-3 Mini (3.8B)	4 GB	★★★☆☆

Conclusion

Self-hosting OpenClaw isn't rocket science – with Docker Compose and Ollama, the stack is up and running in 30 minutes. The key advantage: Full data sovereignty, full GDPR compliance. No data shared with third parties, no DPA debates, no vendor lock-in.

For teams that want to boost productivity with AI without giving up control, self-hosting is the only way.

---

Want to set up OpenClaw on your infrastructure? Talk to us – we help with setup, security hardening, and team training.

More on this topic: What is OpenClaw? · NanoClaw: The lean successor · Our tool philosophy