Ready-to-use OpenClaw agent configurations optimized for local Ollama models. These SOUL.md files use shorter prompts and focused instructions designed for smaller context windows.
| Agent | Model | Use Case |
|---|---|---|
| coding-assistant | gemma4 | Code generation, debugging, refactoring |
| content-writer | qwen3 | Blog posts, social media, marketing copy |
| research-analyst | deepseek-v3 | Topic research, comparisons, analysis |
| project-manager | gemma4:26b | Task planning, PRDs, status updates |
| customer-support | gemma4:e4b | Issue resolution, FAQ, escalation |
- Pull the model you need:
ollama pull gemma4- Copy the SOUL.md to your OpenClaw agent directory:
cp configs/ollama/coding-assistant/SOUL.md ~/.openclaw/agents/coding-assistant/SOUL.md- Launch the agent:
ollama launch openclaw --model gemma4Or run directly with OpenClaw CLI:
openclaw agent --agent coding-assistant --message "Write a Python function to merge two sorted lists"| Hardware | RAM | Recommended Model | Agent |
|---|---|---|---|
| Apple M1/M2 (8GB) | 8 GB | gemma4:e4b | customer-support |
| Apple M1/M2 (16GB) | 16 GB | gemma4 | coding-assistant |
| Apple M3/M4 (16GB+) | 16-32 GB | gemma4:26b | project-manager |
| Desktop GPU (12GB+ VRAM) | 32 GB | deepseek-v3 | research-analyst |
| Any (16GB+) | 16 GB+ | qwen3 | content-writer |
- Token-efficient: Short system prompts that fit in smaller context windows
- Focused rules: Each agent has 5 rules max to avoid instruction drift
- Local-first: No API keys or external dependencies required
- Portable: Just a SOUL.md file — copy it anywhere OpenClaw runs