DEV Community

Cover image for How to Run OpenClaw with Ollama ?
Preecha
Preecha

Posted on

How to Run OpenClaw with Ollama ?

#ai #software #tutorial

TL;DR

Ollama is the easiest way to run powerful AI models locally. Combined with OpenClaw, it creates a free, privacy-focused AI assistant that rivals paid alternatives. This guide walks you through setting up Ollama, choosing the right model, and integrating it with OpenClaw for your personal AI assistant.

Introduction

Running AI locally was once a hobbyist's pursuit, requiring complex setup and expensive hardware. Ollama changed that. With a simple install command and intuitive API, Ollama makes running AI models locally accessible to anyone.

Try Apidog today

ollama launch openclaw --model qwen3.5:35b

When paired with OpenClaw, you get a powerful AI assistant that:

  • Costs nothing to run (after initial setup)
  • Keeps your data 100% private
  • Works offline once models are downloaded
  • Offers full customization control

This guide covers everything you need to get started.

Why Use Ollama with OpenClaw

Benefits of Local AI

  • Privacy: Your conversations never leave your machine
  • No API costs: Pay once for hardware, use unlimited
  • Offline access: Works without internet
  • Full control: Customize models and prompts
  • No rate limits: Use as much as you want

Why Ollama

Ollama stands out for several reasons:

  • Simple installation: One command gets you started
  • Model library: 100+ models available
  • Cross-platform: Works on macOS, Linux, Windows
  • API-first: Easy integration with OpenClaw
  • Active development: Regular updates and new models

Prerequisites

Before starting, ensure you have:

Hardware Requirements

Model SizeMinimum RAMRecommended RAM7B params8GB16GB14B params16GB32GB32B params32GB64GB70B params64GB128GB

Software Requirements

  • macOS 10.15+, Linux, or Windows 10+
  • Administrator/root access for installation
  • Internet connection for initial downloads
  • Command line familiarity

What You'll Need

  • A computer meeting RAM requirements
  • Internet for downloading models
  • Time for initial model downloads (varies by size and connection)

Installing Ollama

macOS Installation

The easiest method uses Homebrew:

brew install ollama
Enter fullscreen mode Exit fullscreen mode

Or use the official installer script:

curl -fsSL https://ollama.ai/install.sh | sh
Enter fullscreen mode Exit fullscreen mode

Linux Installation 

# Using the install script (recommended)
curl -fsSL https://ollama.ai/install.sh | sh

# Or download the binary directly
sudo curl -L https://ollama.ai/download/ollama-linux-amd64 -o /usr/bin/ollama
sudo chmod +x /usr/bin/ollama
Enter fullscreen mode Exit fullscreen mode

Windows Installation

Download the installer Run the installerFollow the on-screen instructions

Download Ollama

Verifying Installation 

ollama --version
Enter fullscreen mode Exit fullscreen mode

You should see output like ollama version 0.15.0 or newer.

Ollama version on terminal

Starting Ollama Service

Ollama runs as a background service:

# Check if Ollama is running
ollama list

# Start Ollama if not running
ollama serve
Enter fullscreen mode Exit fullscreen mode

Check if ollama is running with Ollama list command

Choosing the Right Model

Ollama supports 100+ models. Here's how to choose:

By Use Case

Use CaseRecommended ModelsGeneral conversationQwen3.5, Llama 3.2, MistralCoding assistanceQwen3.5-Coder, DeepSeek-CoderReasoning/mathDeepSeek-R1, Qwen3.5Smaller hardwarePhi3.5, Gemma2.2B

By Hardware

Available RAMRecommended8GB7B models (Qwen3.5, Llama3.2, Mistral)16GB8-14B models32GB14-32B models64GB+70B+ models

Popular Models in 2026

Qwen3.5 — Excellent all-around performance, strong reasoning, good for coding. The most popular choice for OpenClaw in 2026.

DeepSeek-R1 — Open-source reasoning model that rivals GPT-4 on math and logic tasks. Great for complex problem-solving.

Mistral — Lightweight but capable. Excellent for systems with limited RAM.

Installing Models

Pulling Models 

# Install Qwen3.5 (recommended for most users)
ollama pull qwen2.5:7b

# Or the latest Qwen3
ollama pull qwen3:7b

# DeepSeek-R1 for reasoning tasks
ollama pull deepseek-r1:7b

# Llama 3.2
ollama pull llama3.2:7b

# Mistral
ollama pull mistral:7b
Enter fullscreen mode Exit fullscreen mode

Model Tags

Models come in different sizes:

# Different parameter sizes
ollama pull qwen2.5:3b    # Smaller, faster
ollama pull qwen2.5:7b    # Balanced
ollama pull qwen2.5:14b   # More capable
Enter fullscreen mode Exit fullscreen mode

Viewing Installed Models 

ollama list
Enter fullscreen mode Exit fullscreen mode

This shows all downloaded models and their sizes.

Running and Testing Models

Interactive Mode 

# Chat with the model
ollama run qwen2.5:7b
Enter fullscreen mode Exit fullscreen mode

Type your message and press Enter. Type /bye to exit.

API Mode

Ollama runs an API server on port 11434 by default:

# Generate endpoint
curl http://localhost:11434/api/generate -d {
  "model": "qwen2.5:7b",
  "prompt": "Hello, how are you?",
  "stream": false
}
Enter fullscreen mode Exit fullscreen mode

Using the Python Library 

from ollama import Client

client = Client()
response = client.chat(
    model='qwen2.5:7b',
    messages=[
        {'role': 'user', 'content': 'Hello!'}
    ]
)
print(response['message']['content'])
Enter fullscreen mode Exit fullscreen mode

Testing with Apidog

Before connecting to OpenClaw, test your Ollama setup using Apidog:

  • Create a new request in Apidog
  • Set method to POST
  • Enter URL: http://localhost:11434/api/generate
  • Add header: Content-Type: application/json

Create request in Apidog

  1. Add body: 
{
  "model": "qwen3-coder",
  "prompt": "What is 2 + 2?",
  "stream": false
}
Enter fullscreen mode Exit fullscreen mode

Add Body to a request in Apidog

This verifies your Ollama setup works before integrating with OpenClaw.

Integrating Ollama with OpenClaw

Now let's connect Ollama to OpenClaw.

Method 1: Quick Configuration 

# Set OpenClaw to use Ollama with your model
openclaw models set ollama/qwen2.5:7b
Enter fullscreen mode Exit fullscreen mode

Method 2: Environment Variables 

# Configure Ollama endpoint
export OLLAMA_HOST=http://localhost:11434

# Set the default model
export OLLAMA_MODEL=qwen2.5:7b
Enter fullscreen mode Exit fullscreen mode

Method 3: Configuration File

Create or edit ~/.openclaw/config.yaml:

models:
  default: ollama/qwen2.5:7b

ollama:
  host: http://localhost:11434
  model: qwen2.5:7b
  temperature: 0.7
  top_p: 0.9
Enter fullscreen mode Exit fullscreen mode

Verifying Integration 

# Check OpenClaw model status
openclaw models status

# Test with a message
openclaw chat "Hello!"
Enter fullscreen mode Exit fullscreen mode

You should receive a response from your local model.

Configuration Options

Fine-tune your Ollama + OpenClaw setup:

Temperature

Controls creativity vs precision:

ollama:
  temperature: 0.7    # 0.0 = precise, 1.0 = creative
Enter fullscreen mode Exit fullscreen mode

Top-P and Top-K

Control response diversity:

ollama:
  top_p: 0.9         # Nucleus sampling
  top_k: 40          # Token selection
Enter fullscreen mode Exit fullscreen mode

Context Length

For longer conversations:

ollama:
  context_size: 4096  # Default is often 2048 or 4096
Enter fullscreen mode Exit fullscreen mode

System Prompt

Customize model behavior:

ollama:
  system_prompt: |
    You are a helpful coding assistant.
    Provide clear, concise code examples.
    Explain concepts simply.
Enter fullscreen mode Exit fullscreen mode

Switching Between Models

One advantage of Ollama is easy model switching:

# Switch to DeepSeek-R1 for reasoning
openclaw models set ollama/deepseek-r1:7b

# Switch to Qwen-Coder for coding tasks
openclaw models set ollama/qwen2.5-coder:7b

# Switch back to general model
openclaw models set ollama/qwen2.5:7b
Enter fullscreen mode Exit fullscreen mode

Multiple Model Setup

Configure multiple models in config.yaml:

models:
  default: ollama/qwen2.5:7b
  coding: ollama/qwen2.5-coder:7b
  reasoning: ollama/deepseek-r1:7b
Enter fullscreen mode Exit fullscreen mode

Then switch between them:

openclaw models set coding
openclaw models set reasoning
Enter fullscreen mode Exit fullscreen mode

Troubleshooting

Model Won't Load

Problem: Out of memory errors

Solutions:

  • Use a smaller model (7B instead of 14B)
  • Close other applications to free RAM
  • Check available memory with free -h (Linux) or Activity Monitor (Mac)

Slow Responses

Problem: Responses take too long

Solutions:

  • Use a smaller model
  • Enable GPU acceleration (if available)
  • Reduce context size
  • Use SSD storage for model files

Connection Refused

Problem: OpenClaw can't connect to Ollama

Solutions:

# Verify Ollama is running
ollama serve

# Check the port
curl http://localhost:11434
Enter fullscreen mode Exit fullscreen mode

Model Not Found

Problem: Model doesn't exist in Ollama

Solutions:

# Pull the model
ollama pull qwen2.5:7b

# Check available models
ollama list
Enter fullscreen mode Exit fullscreen mode

Conclusion

You've now got a powerful, private AI assistant running locally. Ollama + OpenClaw delivers capabilities that would cost $20+/month with cloud alternatives—all running on hardware you control.

What you can do now:

  • Chat with your AI through multiple platforms
  • Switch between models based on tasks
  • Customize prompts for specialized behaviors
  • Run offline once models are downloaded

The only limit is your hardware.

Next steps:

  • Experiment with different models
  • Try Qwen3.5, DeepSeek-R1, and others
  • Customize your system prompts
  • Explore OpenClaw skills on ClawHub

Ready to build professional AI applications? Download Apidog free and test your AI integrations with a visual interface designed for developers.

FAQ

What's the best Ollama model for OpenClaw?

Qwen3.5 is currently the most popular—balanced performance with good reasoning and coding capabilities. DeepSeek-R1 excels at reasoning tasks if that's your priority.

Can I run multiple Ollama models at once?

Yes, but each model requires RAM. A typical setup runs one model at a time, switching as needed.

Do I need a GPU?

No, Ollama works on CPU. GPU acceleration makes it faster but isn't required. Smaller models (7B) work reasonably well on CPU.

How do I update models? 

ollama pull model-name
Enter fullscreen mode Exit fullscreen mode

Ollama updates automatically if a newer version is available.

Can I use my own fine-tuned models?

Yes, import custom models using Ollama's import functionality. Check the Ollama documentation for details.

Top comments (0)