Bug type

Regression (worked before, now fails)

Summary

Bug Description

The OpenClaw Control UI incorrectly handles provider prefixes when attempting to switch to local models (e.g., Ollama) via the model dropdown, resulting in a GatewayRequestError: model not allowed exception.

Steps to reproduce

Steps to Reproduce

1. Configure openclaw.json with an Ollama auth profile and register a local model in agents.defaults.models (e.g., ollama/qwen2.5:32b or ollama/huihui_ai/qwen2.5-abliterate:7b-instruct).
2. Open the OpenClaw Control UI.
3. Attempt to switch the active model using the UI dropdown to one of the local Ollama models.

Expected behavior

Expected Behavior

The UI should send the exact model string (including the ollama/ prefix) to the Gateway, and the model switch should succeed.

Actual behavior

Actual Behavior

The UI modifies or strips the provider prefix. This results in the Gateway routing the request to the default provider (Google) or failing the allowlist check.
Error examples observed in the UI:

• Failed to set model: GatewayRequestError: model not allowed: google/qwen2.5:32b (UI replaced ollama/ with google/)
• Failed to set model: GatewayRequestError: model not allowed: huihui_ai/qwen2.5-abliterate:7b-instruct (UI stripped the ollama/ prefix)

OpenClaw version

v2026.3.13

Operating system

Windows 11

Install method

npm global

Model

ollama/huihui_ai/qwen2.5-abliterate:7b-instruct

Provider / routing chain

openclaw -> local ollama (http://127.0.0.1:11434)

Additional provider/model setup details

Additional Context

Bypassing the UI and setting the model directly via the backend (session_status tool) works perfectly, confirming that the gateway routing and local configuration are valid. The bug is isolated to how the Control UI frontend constructs the model change request.

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response