openclaw - 💡(How to fix) Fix Feature Request: Notify user on model fallback switch [1 comments, 2 participants]

Problem

When the primary model fails (e.g., Ollama timeout, server unreachable), OpenClaw silently falls back to the configured fallback model. The user is never informed about this switch.

Impact

• User receives responses from a different model without knowing
• Quality/reasoning differences between models can confuse the user
• Silent degradation of experience

Steps to Reproduce

1. Set primary model to e.g. ollama/kimi-k2.6:cloud
2. Set fallback to ollama/glm-5.1:cloud
3. Trigger a failure on primary (disconnect Ollama, wait for timeout)
4. Observe: Agent continues with fallback, user gets no notification

Desired Behavior

Option A: Simple notification

When fallback is activated, send a message to the user:

⚠️ Primary model kimi-k2.6:cloud unavailable. Switched to fallback glm-5.1:cloud.

When primary recovers:

✅ Primary model kimi-k2.6:cloud restored.

Option B: Configurable policy

Add config options:

"agents": {
  "defaults": {
    "model": {
      "primary": "ollama/kimi-k2.6:cloud",
      "fallbacks": ["ollama/glm-5.1:cloud"],
      "notifyOnFallback": true,
      "requireApprovalForFallback": false
    }
  }
}

Option C: Auto-retry with backoff

Instead of permanent switch, retry primary model every N requests and auto-restore.

Environment

• OpenClaw version: 2026.4.15
• Runtime: agent mode
• Primary model: ollama/kimi-k2.6:cloud
• Fallback: ollama/glm-5.1:cloud

Additional Context

This happened today during a long-running embedding task. The agent silently switched to GLM 5.1 without the user knowing, leading to confusion about response quality.

Feature requested by a user via their OpenClaw agent