[PCcoding666]

Summary

Add a mainland Beijing Alibaba Cloud Model Studio chat preset with qwen3.8-max. The setup form offers the model immediately and accepts either the Beijing default URL or a workspace-specific Beijing URL. Saving and reopening a workspace connection keeps its provider label. The English and Chinese guides include both address formats.

Refs #240. This PR covers assistant chat. The issue also tracks other providers and modalities.

Live Beijing acceptance on Apple Silicon

At a63caebd, I started with an empty OpenDirector project and selected qwen3.8-max using a mainland Beijing Model Studio account. OpenDirector’s HTTP assistant and model gateway ran the bundled OpenCode 1.19.5 against the default Beijing endpoint.

1. The model returned a tool call. OpenDirector executed opendirector asset write-text and registered one file, beijing-check.md, containing Beijing model acceptance passed.
2. The next model request included the applied tool receipt. Its asset ID matched the saved file. The model returned five text chunks, which OpenDirector displayed as one assistant text update confirming the write.
3. After restarting the backend and reopening the project in a fresh session, the saved file contents and the assistant reply were both present.

Both streaming requests returned HTTP 200. Provider-reported usage totaled 7,369 tokens across the two requests (7,207 prompt and 162 completion).

Settings and fixture evidence

Chinese setup	English setup

In Chromium I saved the preset, reopened it, added a model ID manually, and selected Qwen in the assistant picker in both languages. These settings runs used a fixture credential. Assistant picker screenshot

The bundled OpenCode 1.19.5 ran through OpenDirector’s HTTP assistant and model gateway against simulated Qwen SSE responses. I checked streamed text, tool calls, reasoning replay and continuation. The tool wrote a text asset to an isolated OpenDirector project. Its ID matched the tool receipt.

Checks and provider verification

• npm run build, npm run check:i18n, and all 768 Node tests passed in a serial run. One browser-host process-exit test failed in the initial parallel run. Its seven file-level tests passed on rerun.
• 61 related Python checks, 25 native runtime unit tests, and four opted-in bundled OpenCode integration cases passed.
• The live test used the default Beijing endpoint. The workspace-specific URL was covered by configuration tests. Release and cross-platform workflows are left to upstream CI.

Provider references are the official model card and OpenAI-compatible endpoint documentation.

Prepared and tested with Codex assistance.

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-11T06:12:16.048037Z	a63caeb	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.