[Feature Request]: Model Switching & Multiple Chats Within Projects in Web Agent Mode

Problem Statement

In the Web Agent mode on chat.z.ai:

1. Model Overuse & Traffic Deadlock: During periods of heavy platform traffic or capacity spikes on a specific model (e.g. GLM-5.3), users receive "model overuse" errors. Currently, the Web Agent interface locks the project to the model chosen at creation with no ability to switch models mid-flight (e.g., falling back to GLM-5.3-Flash).
2. Single Monolithic Conversation Thread: A project is currently constrained to a single continuous chat session. As development progresses, the conversation thread accumulates excessive context history, becoming heavy, slow to render, and costly in token re-tokenization. Users cannot create a clean second or third chat thread that shares the same underlying project workspace and files.

Proposed Solution / Feature Specification

1. In-Project Model Switching:

   • Make the model identifier at the top-left of the Web Agent window (e.g., GLM-5.3) an interactive dropdown selector.
   • Allow users to switch between available models (e.g., GLM-5.3, GLM-5.3-Flash, future variants) within an existing project without losing workspace state or code artifacts.

2. Multiple Chat Threads Per Project:

   • Add a "+ New Chat" / branch conversation button inside the active project interface.
   • Allow multiple conversation threads to operate against the same project file tree, preview window, and deployment state, preventing conversational context bloat.

Community Context & Origin

• Reported by: Fresh on Z.ai Community Discord
• Discord Thread: https://discord.com/channels/1346756824233148527/1547513690465636413
• User Statement:

  "In the Agent section of the web version, when we start a project and begin developing it, we sometimes receive a 'model overuse' error... My first suggestion would be to add a model switching feature... My second suggestion is to allow users to open a new chat within an existing project. Working within a single chat is difficult, as the conversation becomes too heavy over time."

No files, tests, or code entry points are identified; start by reviewing the Web Agent project interface and the linked Discord context. Done means users can switch models within an existing project and create multiple chat threads that share its workspace, files, preview, and deployment state.