How to Integrate Ollama Models in Openclaw

Table of contents: [Show]

Quick Start: Install Ollama

Ollama lets you run large language models locally. This guide shows you how to integrate Ollama models into Openclaw for private, self-hosted AI inference.

Before starting, ensure you have Openclaw deployed on Linux or your preferred platform. For other deployment options, check out our Fly.io deployment guide.

Step 1: Install Ollama and Pull a Model

First, ensure you have Ollama installed from ollama.ai. Then, pull a tool-capable model from your terminal:

ollama pull llama3.3

ollama pull qwen2.5-coder:32b

Step 2: Configure Openclaw

To make an Ollama model your primary agent model, set it in your ~/.openclaw/openclaw.json configuration file:

{
  "agents": {
    "defaults": {
      "model": { "primary": "ollama/llama3.3" }
    }
  }
}

Step 3: Remote Ollama Configuration

Here is an example of an explicit remote configuration:

{
  "models": {
    "providers": {
      "ollama": {
        "baseUrl": "http://ollama-host:11434", 
        "apiKey": "ollama-local",
        "api": "ollama",
        "models": [
          {
            "id": "gpt-oss:20b",
            "name": "GPT-OSS 20B",
            "reasoning": false,
            "input": ["text"],
            "contextWindow": 8192,
            "maxTokens": 81920
          }
        ]
      }
    }
  }
}

Best Practices and Troubleshooting

Local Only: Ollama runs entirely on your machine—no data leaves your network.
Model Size: Larger models need more RAM. llama3.3 requires ~16GB for optimal performance.
GPU Acceleration: Ollama automatically uses CUDA/Metal if available.
Restart Gateway: After editing openclaw.json, restart the gateway: openclaw gateway restart