Add grok-superimage server-side image mode; improve streaming error handling and transient retry logic

[lijirou12]

Motivation

• Introduce a server-controlled image model grok-superimage-1.0 that centralizes n/size/response_format so clients cannot override these for that model.
• Make streaming endpoints more robust by converting transport/exception failures into SSE error payloads for streaming clients.
• Improve token retry behavior to handle transient upstream failures by attempting alternative tokens when sensible.
• Add UI and default config entries for the new superimage server settings.

Description

• Added grok-superimage-1.0 to the model registry in ModelService and a server-side configuration section [superimage] in config.defaults.toml to control n, size, and response_format.
• Enforced server-side image config for grok-superimage-1.0 by adding _superimage_server_image_config() and routing image request handling to use the server config when request.model == 'grok-superimage-1.0' in app/api/v1/chat.py.
• Implemented safe SSE streaming wrappers: _safe_sse_stream() to stream SSE error payloads on exceptions and _streaming_error_response() to return one-shot SSE error responses for early failures in chat.py, and wrapped streaming returns to use these helpers.
• Prevented chat-stream endpoints from emitting intermediate preview/partial images for chat-formatted image streams by filtering partial stages in ImageWSStreamProcessor and ImageStreamProcessor, and ensured a final empty chat chunk and data: [DONE] are emitted when no final chunk was produced.
• Added transient upstream detection transient_upstream() in app/services/grok/utils/retry.py and updated token retry logic in ChatService to try alternative tokens on transient errors when alternatives exist.
• Renamed ambiguous import get_storage to resolve_storage in admin_api/config.py and renamed the endpoint function to get_storage_mode() to avoid naming conflict and improve clarity.
• Adjusted AppChatReverse timeout selection logic to prefer chat.timeout and fall back to video.timeout/image.timeout when unset.
• Updated admin UI app/static/admin/js/config.js, added superimage fields and select controls for size and response_format, and updated README readme.md to document grok-superimage-1.0 behavior.

Testing

• Ran the project's automated test suite with pytest against the modified codebase and existing tests; the test run completed successfully with no failures.
• Exercised streaming image and chat flows in an integration smoke test to verify SSE error payloads are returned on upstream exceptions and that grok-superimage-1.0 uses server-side n/size/response_format (manual integration assertions in CI), which passed.
• Ran static linting/type checks and they completed without new issues.

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
grok2api	Ready	Preview, Comment	Feb 26, 2026 11:43am

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.