Version: e.g. 0.5.x-xxx
0.7.9

Describe the Bug

Model selector changes are not isolated, and may affect other (already running) chats.
(It may be related to use of MCP server in long chats)

Steps to Reproduce

1. Create a chat (CHAT 1) with model selector set to Qwen 3.6 27b (via an Open AI compatible endpoint), let it run with multiple MCP calls
2. While this is running, open a new chat (CHAT 2) set to Qwen 3.6 35b A3B (via an Open AI compatible endpoint), let it run with multiple MCP calls
3. Look in the backend - from the next MCP call, the requests for CHAT 1 are being executed against the model selected in CHAT 2.

Even when I switch back to CHAT 1 is still showing the original model (e.g. Qwen 3.6 27b) but the requests are going to the model for CHAT 2 (e.g. Qwen 3.6 35b A3B)

Screenshots / Logs

Operating System

• MacOS
• Windows
• Linux