Add grok-superimage server-side image flow, robust image retries and SSE error handling

[lijirou12]

Motivation

• Introduce a server-controlled SuperImage model (grok-superimage-1.0) with consistent server-side image generation parameters and safer streaming semantics.
• Make image generation more robust against upstream review/blocking by adding detection, parallel recovery attempts and configurable thresholds.
• Improve streaming reliability by converting transport failures into SSE error events so clients receive structured error payloads instead of raw 5xx breaks.
• Harden a few infrastructure areas: safer config bootstrap/persistence, proxy handling for app-chat, and retry behavior for transient upstream errors.

Description

• Added a new model entry grok-superimage-1.0 in ModelService and a server-side image config loader _superimage_server_image_config() in the chat API to enforce server superimage settings for that model.
• Implemented _safe_sse_stream() and _streaming_error_response() in app/api/v1/chat.py to catch exceptions during streaming and emit SSE error events (uses orjson), and wrapped streaming returns to use the safe wrapper.
• Updated image and image-edit services to use a fallback default for retry.max_retry, propagate tried_tokens, and add recovery logic: additional parallel recovery attempts configured by image.blocked_parallel_attempts and image.blocked_parallel_enabled; image collectors now only accept images meeting final_min_bytes and run extra attempts when finals are insufficient.
• Enhanced WebSocket reverse (ws_imagine) with blocked-detection tuning (blocked_grace_seconds), parallel retry collection when blocked is suspected, and clearer logging; adjusted classification rules to use a byte-size threshold for final images.
• Modified image stream processors (ImageWSStreamProcessor, ImageWSCollectProcessor, ImageStreamProcessor) to only expose final results for chat-formatted streaming, ensure a final empty chat chunk if none were emitted, and preserve original image names for grok-superimage-1.0 URL outputs (no -final suffix).
• Added transient upstream detection in app/services/grok/utils/retry.py (transient_upstream) and use it in ChatService to retry with alternate tokens on likely transient errors; also defaulted retry.max_retry fallback to 3 where used.
• Improved app chat proxy handling in app/services/reverse/app_chat.py by normalizing SOCKS schemes and using the proxy parameter for SOCKS proxies while using proxies for HTTP proxies, plus better timeout source selection.
• Made config bootstrap safer in app/core/config.py to avoid overwriting remote config with empty data during initialization and log when skipping persistence.
• UI and defaults: added new config keys and UI support for image.blocked_grace_seconds, image.blocked_parallel_attempts, image.blocked_parallel_enabled, and a new [superimage] section in config.defaults.toml; updated readme.md and admin JS (app/static/admin/js/config.js) to expose the new options and document grok-superimage-1.0 behavior.
• Minor renames and imports: get_storage import renamed to resolve_storage in admin config to avoid name clash with endpoint function.

Testing

• Ran the project's automated test suite with pytest (existing tests) and the suite completed without failures.
• Executed a local smoke test of the POST /v1/chat/completions flow for grok-superimage-1.0 including streaming and non-streaming image responses and observed correct server-controlled n/size/response_format and final-only streaming behavior.
• Performed integration checks for WebSocket image generation against the reverse ws_imagine path to validate blocked-detection and parallel recovery logic under simulated preview-only responses.
• Built and exercised the admin UI (npm build/static serve) to confirm new config fields are visible and editable in the admin panel.

---

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 27, 2026 6:21am

[chatgpt-codex-connector]

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