Bug type

Regression (worked before, now fails)

Summary

Bug Description

When selecting a non-Anthropic model from the model dropdown in the dashboard,
the gateway receives the wrong provider prefix (anthropic/) instead of the
correct one.

Environment

• OpenClaw version: 2026.3.13 (61d171a)
• OS: macOS (Apple Silicon, M4 Mac mini)
• Install method: npm (sudo npm install -g openclaw)

Steps to Reproduce

1. Configure both Anthropic and Google (Gemini) models in openclaw.json
2. Open the dashboard (openclaw dashboard)
3. Click the model dropdown
4. Select google/gemini-2.5-flash

Expected Behavior

Gateway receives: google/gemini-2.5-flash

Actual Behavior

Gateway receives: anthropic/gemini-2.5-flash

Error Message

Failed to set model: GatewayRequestError: model not allowed: anthropic/gemini-2.5-flash

Additional Notes

• Same issue occurs with lmstudio/qwen3.5-4b-mlx → becomes anthropic/qwen3.5-4b-mlx
• The config file (openclaw.json) is correct with proper prefixes
• The models work correctly when called directly via API
• The bug appears to be in the dashboard UI which hardcodes anthropic/ prefix
  regardless of the selected model's actual provider

Steps to reproduce

1. Configure both Anthropic and Google (Gemini) models in openclaw.json
2. Open the dashboard (openclaw dashboard)
3. Click the model dropdown
4. Select google/gemini-2.5-flash

Expected behavior

Gateway receives: google/gemini-2.5-flash

Actual behavior

Gateway receives: anthropic/gemini-2.5-flash

OpenClaw version

2026.3.13 (61d171a)

Operating system

macOS (Apple Silicon, M4 Mac mini)

Install method

npm (sudo npm install -g openclaw)

Model

google/gemini-2.5-flash

Provider / routing chain

OpenClaw dashboard UI → OpenClaw gateway (local, port 18789) → Google Generative AI API

Additional provider/model setup details

Google Gemini configured in openclaw.json under models.providers.google with api: "google-generative-ai".
Also reproduced with lmstudio/qwen3.5-4b-mlx (local LM Studio model).
Dashboard sends anthropic/ prefix for all non-Anthropic models.

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response