openclaw with ollama (Zero cost AI Assistant)

openclaw_with_ollama.md

OpenClaw configuration with Ollama

Install OpenClaw

curl -fsSL https://openclaw.ai/install.sh | bash

Install Ollama

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

Run a powerful model locally

ollama run gpt-oss:20b

Uninstall OpenClaw (if exists)

openclaw uninstall

Prepare OpenClaw config for Ollama

{
  "models": {
    "providers": {
      "ollama": {
        "baseUrl": "http://localhost:11434/v1",
        "apiKey": "ollama-local",
        "api": "openai-completions",
        "models": [
          {
            "id": "gpt-oss:20b",
            "name": "gpt-oss:20b",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 131072,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "ollama/gpt-oss:20b"
      },
      "workspace": "/Users/abhi/.openclaw/workspace",
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      }
    }
  },
  "tools": {
    "web": {
      "search": {
        "enabled": false
      },
      "fetch": {
        "enabled": true
      }
    }
  }
}

Start OpenClaw

openclaw onboard

Configure it with Telegram (optional)

Steps explained in the video.

Prompt injections, for one. Frontier models are hardened against this in training, whilst local models tend to rely more on system prompts,which is less secure.

Is the quality of local models (like gpt-oss:20b) enough for everyday agent use? I think they are very poor and stupid.

Hard coding models disables auto discovery.

https://chatgpt.com/share/69968599-8438-800c-81a6-be7a29fd0240

i dont think local model like gpt-oss:20b para will work for all the needs like we wont get that experience like all those when we use claude,gpt models but somehow secutity wise its ok but performence wise we need to search alternative on local machines can run.

Is the quality of local models (like gpt-oss:20b) enough for everyday agent use? I think they are very poor and stupid.

hi, no it sucks.
My Agents always forgets stuff, answers me in different languages and mentions the memory.md all the time

Not hard to add prompt injection defense to local model https://github.com/awiseguy88/openclaw-advanced-prompt-injection-defense-system.git