OpenClaw with Ollama local deployment, super simple!

Are you still worried about the cost of calling a big cloud model? Still worried about data privacy?

I'll teach you how to use it today. OpenClaw + Ollama Create a fully localized AI assistant, all data running locally, safe and cost-effective!

What's Ollama?

Ollama is a tool for you to run large language models locally. It supports many open-source models such as Llama 2, Llama 3, Mistral, Gemma, which can be run without GPU.

Official website: https://ollama.com

Preparations

1. Installation of Ollama

MacOS/ Linux:

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

Windows:
Go straight to https://ollama.com to download the installation package

Download models

After installation, download the model you need:

# 推荐：下载 Llama 3（8B 参数，一般配置够用）
ollama pull llama3

# 或者下载 Mistral（性能不错）
ollama pull mistral

# 配置低的可以选 qwen（阿里开源，中文效果好）
ollama pull qwen:7b

3. Validation of Ollama run

# 启动 Ollama 服务（一般安装后自动启动）
ollama serve

# 测试模型
ollama run llama3

Okay!

OpenClaw Configuration

1. Find Profile

OpenClaw profile in ~/.openclaw/openclaw.json

2. Add Ollama as Custom

Add the following to the profile:

{
  models: {
    providers: {
      ollama: {
        type: "openai",  // Ollama 兼容 OpenAI API
        baseUrl: "http://localhost:11434/v1",
        api: "key",
        defaultHeaders: {
          "Content-Type": "application/json"
        }
      }
    }
  },
  agents: {
    defaults: {
      model: {
        primary: "ollama/llama3"
      },
      models: {
        "ollama/llama3": {
          // 支持的参数
        },
        "ollama/mistral": {},
        "ollama/qwen:7b": {}
      }
    }
  }
}

3. Examples of full configuration

{
  models: {
    providers: {
      ollama: {
        type: "openai",
        baseUrl: "http://localhost:11434/v1"
      }
    }
  },
  agents: {
    defaults: {
      model: {
        primary: "ollama/llama3",
        fallbacks: ["ollama/mistral"]
      },
      models: {
        "ollama/llama3": {
          description: "Meta Llama 3, 8B 参数"
        },
        "ollama/mistral": {
          description: "Mistral 7B, 性能强劲"
        },
        "ollama/qwen:7b": {
          description: "阿里 Qwen 7B, 中文效果好"
        }
      }
    }
  }
}

4. Restart OpenClaw

# 重启 Gateway
openclaw gateway restart

# 查看模型状态
openclaw models status

Common problems

Q1: What about slow-loading?

The initial loading of the model will be slower because of the need to download the model file. You can keep Ollama running in the backstage, avoiding every restart.

# 让 Ollama 保持运行
nohup ollama serve &

Q2: What if there's not enough?

Select the smeller model:
– I'm sorry. llama3:8b ~ Need about 8GB visible
– I'm sorry. mistral:7b ~ Need about 8GB visible
– I'm sorry. qwen:7b ~ Need about 8GB visible
– I'm sorry. llama3:70b ~ It takes about 40 GB to be visible.

Q3: How do we look at the models?

ollama list

Q4: Slow response?

Local model response rates depend on:
– Model size
– Visibility/RAM speed
– CPU Performance

Quantified models recommended (band) :q4_K_M Wait, wait, wait.

Q5: OpenClaw cannot find a model?

Ensure that:
Ollama is running:ollama serve
2. The configured model has been downloaded:ollama list
3. BaseUrl Correct:http://localhost:11434/v1

Progress: Multimodel switching

Configure multiple fallbacks:

{
  agents: {
    defaults: {
      model: {
        primary: "ollama/llama3",
        fallbacks: ["ollama/qwen:7b", "ollama/mistral"]
      }
    }
  }
}

So if llama3 is not available, it will automatically switch to the next model.

Summary

Through Ollama local deployment, you can:

• Zenium Cost savings: Total free, no API Key
• Zenium Privacy: Data not available locally
• Zenium Customise: Free choice of model
• Zenium More stable.: Not dependent on the network

Go ahead and try. Ask me any questions.

---

Make friends, add me to the group, and share A.I. and cyber security technology.