Run OpenClaw 100% Locally with Ollama

Want to run your OpenClaw agent completely offline with zero monthly costs and zero data leaving your VPS? This guide shows you exactly how to switch to **100% local models** using Ollama — and highlights the brand-new **Gemma 4** family that’s currently dominating local performance.

Table of Contents

• 1. Why Run OpenClaw Locally?
• 2. Step-by-Step Setup on Your VPS
• 3. Best Local Models for OpenClaw in 2026 (Gemma 4 Focus)
• 4. How to Switch to Local Models
• 5. Performance Tips for VPS
• 6. Local vs Cloud — When to Use Which
• 7. Gemma 4 & Ollama: The Ultimate CLI Cheat Sheet

1. Why Run OpenClaw Locally?

• ✅ Zero cost after setup (no API bills)
• ✅ 100% private — nothing leaves your VPS
• ✅ No rate limits or usage caps
• ✅ Works offline
• ✅ Full control over your data

2. Step-by-Step Setup on Your VPS

Step 1: Install Ollama

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

Step 2: Pull the Gemma 4 models

ollama pull gemma4:26b        # My current daily driver (excellent balance)
ollama pull gesta4:31b        # Maximum intelligence (if you have the RAM)
ollama pull gemma4:e4b        # Lightweight & fast for lighter tasks

Step 3: Verify Ollama is running

ollama list

3. Best Local Models for OpenClaw in 2026 (Gemma 4 Family)

4. How to Switch OpenClaw to Local Models

Once Ollama is running, switch your agent with one command (this is what I use daily):

openclaw models set ollama/gemma4:26b

Other useful commands:

• openclaw models list --local → See all your Ollama models
• openclaw models status → Check current model + speed
• openclaw models set ollama/gemma4:31b → Switch to the bigger beast when needed

openclaw gateway restart

5. Performance Tips for VPS

• Use at least 3 GB RAM for gemma4:26b (64 GB+ for 31B)
• Enable GPU if your VPS provider offers it (huge speed boost)
• Run ollama serve in the background with systemd
• Keep /think low or medium for faster responses with Gemma 4

6. Local vs Cloud — When to Use Which

openclaw models fallbacks add venice/claude-4.6-opus

7. Gemma 4 & Ollama: The Ultimate CLI Cheat Sheet

Here’s a clean, practical list of Ollama CLI commands specifically for working with Gemma 4 (and general Ollama usage). All commands are run in your terminal.

1. Download / Pull Gemma 4 Models

ollama pull gemma4          # Default (E4B, ~9.6 GB) – recommended starting point
ollama pull gemma4:e2b      # Smaller & faster (~7.2 GB, great for laptops)
ollama pull gemma4:e4b      # Same as default but explicit tag
ollama pull gemma4:26b      # MoE model (~18 GB)
ollama pull gemma4:31b      # Largest dense model (~20 GB)

2. Run & Chat with Gemma 4

ollama run gemma4           # Starts interactive chat with default model
ollama run gemma:e2b       # Or any specific tag

One-shot prompts (run once without entering chat):

ollama run gemma4 "Write a Python script to scrape a webpage"
ollama run gemma4:e2b "Explain quantum computing in simple terms"

3. Multimodal (Image + Text) – Gemma 4 supports vision!

Put the image path at the end of the prompt:

ollama run gemma4 "Describe this image in detail" /path/to/your/photo.jpg
ollama run gemma4 "What’s written on this document?" ~/Desktop/invoice.png

(Works with the e2b/e4b variants too — they even support audio.)

4. Model Management Commands

ollama list                 # (or ollama ls) → see all models you have downloaded
ollama ps                   # show currently running models
ollama show gemma4          # view model info (parameters, architecture, etc.)
ollama show --modelfile gemma4   # see the full Modelfile
ollama rm gemma4            # delete a model to free up space
ollama cp gemma4 my-gemma4  # make a copy/rename
ollama stop gemma4          # stop a running model

5. Useful Flags & Extras

• Start the Ollama server in the background:
  

  ollama serve

• Keep the model in memory longer (useful for repeated use):
  

  ollama run gemma4 --keepalive 30m

• Run with custom temperature (more creative or more deterministic):
  

  ollama run gemma4 --temperature 0.7

6. Quick Commands Inside an ollama run Session

Once you’re in the chat (after ollama run gemma4), you can type these:

• /bye or /exit → quit
• /clear → clear chat history
• /system You are a world-class Python developer. → set a new system prompt
• /temperature 0.8 → change temperature on the fly

That’s it! You now have a completely private, zero-cost OpenClaw agent running 24/7 on your VPS — powered by the latest Gemma 4 models.