How to Run OpenClaw on a Home Server Using Docker and Ollama

Hosting your own AI agent is the ultimate power move for privacy and cost savings in 2026. OpenClaw (formerly Clawbot) has become the gold standard for "agentic" workflows—allowing an AI to control your computer, manage your calendar, and even send messages via WhatsApp or Telegram—all from your own hardware.

By combining Docker for containerization and Ollama for model serving, you create a stable, isolated environment that can run 24/7. (Be sure to review how doing this saves you $15,000/year).

1. Home Server Minimum Requirements (2026)

Running an agent is more taxing than a simple chatbot because the model must remain "resident" in memory to respond to background triggers. This heavy VRAM requirement dictates your build.

Check out our Hardware Selection Page for deep dives on specific 2026 GPU benchmarks.

2. The Step-by-Step Setup

Step A: Install the Foundation

Before touching OpenClaw, you need the "engine" (Ollama) and the "container" (Docker).

• Install Ollama: Download from ollama.com.
• Install Docker: Download Docker Desktop for Windows/Mac or use apt install docker.io for Linux.

Step B: Pull Your AI Model

For OpenClaw to be effective, you need a model that is excellent at Tool Calling. In 2026, the current "Sweet Spot" model is Qwen 3 Coder. Open your terminal and run:

ollama pull qwen3-coder:32b

Visit our Model Rankings Page to see why we prefer Qwen 3 over Llama 3 for agentic tasks.

Step C: Deploy OpenClaw via Docker

OpenClaw provides an official Docker Compose file that simplifies the networking between the agent and your local files.

1. Clone the Repo:
   git clone https://github.com/openclaw/openclaw && cd openclaw
2. Configure the Environment:
   Create a .env file in the folder. If you are using Ollama locally, set your provider:

   OPENCLAW_LLM_PROVIDER=ollama OLLAMA_BASE_URL=http://host.docker.internal:11434

3. Launch:
   docker-compose up -d

3. PC vs. Mac: Key Differences

On Windows (PC):

• Networking: In your .env file, you must use host.docker.internal to allow the Docker container to talk to the Ollama service running on your Windows host.
• WSL2: Ensure Docker is set to use the WSL2 backend for significantly faster file access when the agent is writing code.

On macOS (Mac Studio/Mini):

• Unified Memory: If you are on an M2/M3/M4 Ultra, you don't need to worry about VRAM vs System RAM. Ollama will automatically utilize your unified memory.
• Permissions: macOS will prompt you for "Full Disk Access." For OpenClaw to manage your files or calendar, you must grant this to the Docker process under System Settings > Privacy & Security.

4. Final Onboarding

Once the container is running, access the OpenClaw TUI (Terminal User Interface) to link your messaging apps:

docker exec -it openclaw-cli openclaw onboard

The wizard will guide you through:

• Linking WhatsApp/Telegram: Scan the QR code that appears in your terminal.
• Defining Skills: Enable the "Filesystem" and "Web Browser" skills so your agent can actually perform tasks.

5. Troubleshooting Common OpenClaw & Docker Errors

Even with a perfect setup, local AI in 2026 has its quirks. If your agent isn't responding or you see "red text" in your terminal, check these verified fixes.

The "Fetch Failed" / Connection Refused Error

The Symptom: You see TypeError: fetch failed or ECONNREFUSED 127.0.0.1:11434 in the Docker logs.

The Cause: Inside a Docker container, 127.0.0.1 refers to the container itself, not your PC where Ollama is running.

The Fix (Mac/Windows): Change your OLLAMA_BASE_URL in the .env file to http://host.docker.internal:11434.

The Fix (Linux): Docker for Linux doesn't support host.docker.internal by default. Add this to your docker-compose.yml under the openclaw service:

extra_hosts:
  - "host.docker.internal:host-gateway"

The "Pairing Required" Loop

The Symptom: The dashboard says "Error: pairing required" or "1008: pairing required."

The Cause: OpenClaw requires explicit admin approval for any new "device" (even your own browser) to prevent unauthorized access.

The Fix:
1. Run this command in your terminal: docker exec -it openclaw-gateway openclaw devices list
2. Find your Request ID and run: docker exec -it openclaw-gateway openclaw devices approve <REQUEST_ID>

WhatsApp "Logging In" Hangs

The Symptom: You scan the QR code, your phone says "Linking," but the OpenClaw dashboard stays stuck on "Logging in..."

The Cause: This is a known 2026 bug with the session handshake.

The Fix: Restart the container while the phone says "Logging in."
docker-compose restart openclaw-gateway
Wait 30 seconds, and the session should finalize. If it fails, clear the cache with rm -rf ~/.openclaw/credentials/whatsapp and try again.

"Truncating Input Messages" (The Context Wall)

The Symptom: The agent "forgets" the beginning of the conversation or gives short, halluncinated answers.

The Cause: Ollama defaults to a 4,096 (4k) context window, which is too small for agents.

The Fix: You must manually increase the context in your Ollama Modelfile.

FROM qwen3-coder:32b
PARAMETER num_ctx 65536

Then run: ollama create qwen3-agent -f Modelfile and update your OpenClaw config to use the new qwen3-agent model.

6. Security Warning: Don't Be a Statistic

OpenClaw has "omnipotent" control over your system.

• NEVER expose port 18789 (the gateway) to the public internet without a VPN like Tailscale.
• USE A SANDBOX: If you are running untrusted "Skills," run OpenClaw in a restricted Docker volume so it cannot delete your actual PC files.
• MONITOR LOGS: Occasionally run docker compose logs -f to see if your agent is trying to execute unexpected commands.

Common Questions & Solutions

Q: I’m on Windows/Mac. Why is my GPU not being used?

The Fix: If you run Ollama inside Docker, you need to install the NVIDIA Container Toolkit (for Windows/Linux) or ensure Docker Desktop has "GPUS" enabled. However, for most home servers, it is highly recommended to run Ollama natively on the host OS and use the host.docker.internal link provided in the script above. This avoids the massive performance overhead of virtualized GPU drivers.

Q: How do I update OpenClaw?

The Fix: Unlike a standard app, you don't "click update." Run these three commands in your folder:

docker compose pull
docker compose up -d --remove-orphans
docker exec -it openclaw-gateway openclaw doctor --fix

(This ensures your old config files match the new version's requirements).

Q: The agent says "I can't see those files."

The Fix: Remember that OpenClaw is "sandboxed." It can only see files inside the ~/openclaw/workspace folder on your computer. If you want it to see your Documents or Downloads, you must add them as additional volumes in the docker-compose.yml file, like this:
- ~/Downloads:/home/node/.openclaw/workspace/downloads

Q: My server restarted and OpenClaw didn't come back.

The Fix: Ensure your Docker service is set to start on boot. On Linux, run sudo systemctl enable docker. The restart: unless-stopped line in our script handles the rest.