🤖 AI Agents

This isn’t a post about how successful AI is. It’s more about my hands‑on experience running local LLMs on my new Mac mini. 🖥️✨

🐾 Openclaw

The first thing I wanted to try on my new machine was Openclaw—it’s been everywhere lately. Installing and configuring it was surprisingly easy. Just follow the instructions on the official website. 🚀

I’m using Ollama to run local models. Openclaw recommends glm-4.7-flash:latest, but that one doesn’t run on my device, so I switched to qwen3:8b.

Before using it, I had to increase the default context length, which was way too small. Here’s what I did:

1. Create a Modelfile

   echo "FROM qwen:8b" > Modelfile
   echo "PARAMETER num_ctx 32768" >> Modelfile

2. Create the new model

   ollama create qwen-claw -f Modelfile

3. Run configuration

   openclaw configure

However… one simple “hello” takes almost 2 minutes to respond on telegram. 🐌💬
A remote model is definitely needed for a smoother experience.

🧩 Opencode

My experience with Opencode was quite different. I tested two models:

• qwen2.5-coder:7b
• qwen3:8b

Before starting, set environment variable OLLAMA_HOST to 0.0.0.0 and you need to manually edit the config file ~/.config/opencode/opencode.json.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (Local)",
      "options": {
        "baseURL": "http://<WSL Host>:11434/v1"
      },
      "models": {
        "qwen2.5-coder:7b": {
          "name": "qwen2.5-coder:7b"
        },
        "qwen3:8b": {
          "name": "qwen3:8b"
        }
      }
    }
  }
}

Results weren’t great:

• qwen2.5-coder hallucinated and got stuck in tool‑calling loops 🤯🔧
• qwen3 kept repeating itself endlessly 🔁😵

Not ideal for coding tasks at the moment.