openclaw - 💡(How to fix) Fix Existing channel sessions remain pinned to old model after default model changes

This was generated by AI during triage.

Bug: existing channel sessions stay pinned to old model after default model changes, and provider baseUrl may diverge between config files

Summary

After changing the default model to tencentcodingplan/glm-5, an existing QQ direct session continued using deepseek/deepseek-v4-pro. The session failed because the DeepSeek key had insufficient balance, even though openclaw models status --json showed tencentcodingplan/glm-5 as the configured default.

There was also a provider URL mismatch: the main config had Tencent Coding Plan configured as https://api.lkeap.cloud.tencent.com/coding/v3, while the agent-level models.json had https://api.lkeap.cloud.tencent.com/coding/v3/v1.

Environment

• OpenClaw version: OpenClaw 2026.5.2 (8b2a6e5)
• Gateway port: 15100
• Channel involved: qqbot
• Default model reported by openclaw models status --json: tencentcodingplan/glm-5

What happened

1. Default model was configured as:

   {
     "defaultModel": "tencentcodingplan/glm-5",
     "resolvedDefault": "tencentcodingplan/glm-5"
   }

2. The existing QQ direct session still had persisted session-level model state:

   {
     "key": "agent:main:qqbot:direct:<redacted>",
     "modelProvider": "deepseek",
     "model": "deepseek-v4-pro",
     "systemPromptReport": {
       "provider": "deepseek",
       "model": "deepseek-v4-pro"
     }
   }

3. New messages in that QQ direct session continued to use DeepSeek and failed with:

   deepseek (deepseek-v4-pro) returned a billing error
   402 Insufficient Balance

4. Manually changing the session entry to:

   {
     "modelProvider": "tencentcodingplan",
     "model": "glm-5",
     "systemPromptReport": {
       "provider": "tencentcodingplan",
       "model": "glm-5"
     }
   }

   made the same QQ direct session work again.

5. Validation after the manual fix:

   winnerProvider: tencentcodingplan
   winnerModel: glm-5
   fallbackUsed: false

Provider URL mismatch

The main config contained:

{
  "models": {
    "providers": {
      "tencentcodingplan": {
        "baseUrl": "https://api.lkeap.cloud.tencent.com/coding/v3"
      }
    }
  }
}

But the agent-level model file contained:

{
  "providers": {
    "tencentcodingplan": {
      "baseUrl": "https://api.lkeap.cloud.tencent.com/coding/v3/v1"
    }
  }
}

Changing the agent-level file back to https://api.lkeap.cloud.tencent.com/coding/v3 fixed the mismatch.

Expected behavior

• If the default model changes, OpenClaw should make it clear when existing sessions remain pinned to an older model.
• A failed session pinned to a non-default provider should be discoverable via doctor, models status, or a sessions command.
• There should be a supported CLI way to clear or update a session-level model override.
• Provider baseUrl values should not silently diverge between the main config and agent-level models.json, and OpenClaw should not accidentally append /v1 when the configured OpenAI-compatible endpoint already includes its required path.

Suggested fixes

• Add a command such as:

  openclaw sessions clear-model <session>
  openclaw sessions set-model <session> <provider/model>

• Have openclaw doctor report sessions whose pinned model differs from the current default, especially failed/recent sessions.

• Clarify in openclaw models status whether the default model only applies to new sessions, and whether any active sessions have overrides.

• Ensure provider URL normalization does not append /v1 to Tencent Coding Plan's https://api.lkeap.cloud.tencent.com/coding/v3 endpoint.

Workaround

Manually edit ~/.openclaw/agents/main/sessions/sessions.json and update the affected session's modelProvider, model, and systemPromptReport fields, then verify with:

openclaw sessions --json --active 20
openclaw agent --session-id <session-id> --message ping --timeout 60 --json