Bug type

Regression (worked before, now fails)

Summary

Bug Description

When switching models from the WebUI / Control UI chat header picker, the selected model can be sent as a bare model id instead of a full provider/model reference.

If the current session is already pinned to another provider, the gateway resolves the bare model id against that current/default provider and rejects the request with an error like:

GatewayRequestError: model not allowed: openai-codex/glm-4.7

This makes valid models from other providers impossible to select from the picker.

Steps to Reproduce

1. Open the WebUI / Control UI.
2. Use a session currently associated with openai-codex/gpt-5.4 or another provider-specific model.
3. Open the chat header model picker.
4. Select a model that belongs to another provider, for example glm-4.7.
5. Observe that the request fails with a model not allowed error using the wrong provider prefix.

Expected Behavior

The model picker should submit the selected model as a full provider/model reference based on the catalog entry shown in the picker.

Selecting glm-4.7 should resolve to its configured provider, not inherit the current session provider.

Environment

• OpenClaw WebUI / Control UI
• Reproduced on March 15, 2026
• Session currently pinned to openai-codex/gpt-5.4
• Another provider exposes glm-4.7

Root Cause

The chat header model picker stored entry.id as the <option> value and only used the provider for the visible label.

When the selection changed, WebUI sent sessions.patch({ model: "<bare-model-id>" }) instead of sessions.patch({ model: "<provider>/<model>" }).

Proposed Fix

• Build picker option values from the catalog's full provider/model identity
• Preserve the provider in the locally cached selected value
• Add a regression test to ensure the picker submits the full model reference

Steps to reproduce

1. open webui - chat
2. change to another provider/model, like openai to glm
3. error shows

Expected behavior

no error

Actual behavior

OpenClaw version

3.14 source code (commit 8aaafa0)

Operating system

Macos 26.3

Install method

pnpm dev

Model

openai-codex / glm

Provider / routing chain

openai-codex

Config file / key location

No response

Additional provider/model setup details

No response

Logs, screenshots, and evidence

03:50:29 info gateway/ws {"subsystem":"gateway/ws"} ⇄ res ✗ sessions.patch 29ms errorCode=INVALID_REQUEST errorMessage=model not allowed: openai-codex/glm-4.7 conn=f1084474…4add id=8c3f5cc6…7404

Impact and severity

No response

Additional information

No response