lmstudio-model-switch（lmstudio 模式切换）

lmstudio-model-switch

Fast model switching between LM Studio local and Kimi API for OpenClaw.

Switch your agent's AI model on-the-fly between local (LM Studio) and cloud (Kimi API) providers with a single command.

Installation

# Clone to your OpenClaw skills directory
git clone https://github.com/yourusername/lmstudio-model-switch \
  ~/.openclaw/workspace/skills/lmstudio-model-switch

# Or manually copy
cp -r lmstudio-model-switch ~/.openclaw/workspace/skills/

Usage

Commands

Examples

# Check current status
/switch-model status

# Switch to local LM Studio
/switch-model local

# Switch to specific model
/switch-model local mistral-small-24b

# Switch to Kimi API
/switch-model api

Configuration

Add to your openclaw.json:

{
  "skills": {
    "lmstudio-model-switch": {
      "enabled": true,
      "config": {
        "local": {
          "baseUrl": "http://127.0.0.1:1234/v1",
          "defaultModel": "qwen/qwen3.5-9b"
        },
        "api": {
          "provider": "kimi-coding",
          "model": "k2p5"
        }
      }
    }
  }
}

How It Works

1. Backup: Creates timestamped backup of openclaw.json
2. Edit: Modifies "primary" model in agents.defaults
3. Verify: Validates JSON syntax
4. Restart: Restarts OpenClaw gateway service
5. Confirm: Reports new active model

Use Cases

Privacy-First Work

Use local when handling:

• Authentication tokens
• Passwords or credentials
• Sensitive personal data
• Proprietary code

Quality-First Work

Use API when needing:

• Maximum reasoning capability
• Very long contexts (>100k tokens)
• Best-in-class performance
• Cloud reliability

VRAM Management

Switch to API when:

• GPU memory is low (<6GB free)
• Running other GPU-intensive tasks
• LM Studio is restarting

Requirements

• OpenClaw ≥ 2026.3.12
• LM Studio running on port 1234 (for local mode)
• Kimi API key configured (for API mode)
• systemd (for service restart)

Troubleshooting

"LM Studio not responding"

# Check if LM Studio is running
curl http://127.0.0.1:1234/api/v0/models

# Restart LM Studio if needed
killall lmstudio; sleep 2; lmstudio &

"Switch failed"

• Check JSON syntax: python3 -m json.tool ~/.openclaw/openclaw.json
• Restore from backup: cp ~/.openclaw/openclaw.json.bak.* ~/.openclaw/openclaw.json

"Gateway won't restart"

# Check service status
systemctl --user status openclaw-gateway

# Manual restart
systemctl --user restart openclaw-gateway

Author

WarMech - OpenClaw Community

License

MIT

Changelog

2026-03-14 - v1.0.0

• Initial release
• Support for local/API switching
• Backup/restore functionality
• systemd integration