- name
- model-switcher
- description
- Switch the OpenClaw default model safely. Triggers: switch model, change model, 切换模型, 换模型, switch to XX model. Use when the user wants to change the default AI model. Reads openclaw.json, validates the target model exists in the configured provider list, backs up config, modifies only agents.defaults.model.primary, restarts gateway, and auto-rolls back on failure.
Model Switcher
Switch the OpenClaw default model with backup, validation, and automatic rollback.
Workflow
Step 1: Validate Model Name
Read ~/.openclaw/openclaw.json. Extract:
- Current model from
agents.defaults.model.primary - Available models from
models.providerslist
Match user request against available models (exact, alias, or partial match):
- Exact:
zai/glm-5→ direct hit - Alias: "GLM 5" →
zai/glm-5; "doubao" →volcengine-plan/doubao-seed-2.0-pro - Partial: "deepseek" →
volcengine-plan/deepseek-v3.2
Step 2: Abort if Not Found
If no match, stop and report:
- List all available models
- Ask user to pick one
Never guess or execute blindly.
Step 3: Execute Switch
python3 scripts/model_switch.py <full-model-name> --retry 3The script handles: backup → modify config → restart gateway → health check → auto-rollback on failure.
Step 4: Expect Session Interruption
Gateway restart kills the current session. This is normal. The session resumes automatically after Gateway recovers.
Constraints
- Only modify
agents.defaults.model.primary— never touch the provider list - Always validate model exists before switching
- Backups stored in
~/.openclaw/config_backups/ - Script has full backup + rollback — config will never be lost