openclaw - 💡(How to fix) Fix WebUI model picker sends model without provider prefix, causing model not allowed in enterprise LiteLLM setups [4 comments, 2 participants]

Problem

When using OpenClaw with an enterprise LiteLLM proxy that manages multiple external LLM providers (e.g., bailian, qianfan), users cannot persist a default model selection via WebUI.

Scenario

models.providers.litellm.models:
  - bailian/glm-5      # External provider A
  - qianfan/glm-5     # External provider B
  - bailian/qwen3.5-plus

Expected: User selects bailian/glm-5 in WebUI → sends litellm/bailian/glm-5 to backend → matches agents.defaults.models → allowed

Actual: User selects bailian/glm-5 in WebUI → sends bailian/glm-5 (without litellm/ prefix) → does NOT match agents.defaults.models (which contains litellm/bailian/glm-5) → rejected as "model not allowed"

Why This Matters for Enterprise

Enterprise LiteLLM deployments typically:

1. Use LiteLLM as a unified gateway with centralized API key management
2. Want to expose real provider/model names to end users for transparency
3. Need to control which models users can access via agents.defaults.models
4. Must NOT expose provider API keys directly to end users

The current behavior breaks the enterprise pattern where:

• Provider prefix (litellm/) = internal routing layer
• Model suffix (bailian/glm-5) = what users see and select

Expected Behavior

The WebUI model picker should send the full model key (e.g., litellm/bailian/glm-5) that matches agents.defaults.models, OR the backend should accept both provider/model and full-prefix/provider/model formats.

Alternative Workaround

If agents.defaults.models is left empty, WebUI shows all models, but users cannot persist any default model selection.