Greptile Summary

This PR raises the per-image upload limit from 4 MB to 50 MB (matching the Azure OpenAI recommendation), increases the maximum image count from 10 to 16, and recalculates maxImageBodySize accordingly (~816 MB). A regression test covering the previously-rejected 5 MB image is also added.

• defaultMaxImageFileSize bumped to 50 MB and maxImageCount to 16; maxImageBodySize is now a derived formula rather than a hand-coded constant.
• buildMultipartJSONBody base64-encodes every image for tracing — at the new limits (16 × 50 MB) this can produce ~2.7 GB in-process per request, which is a meaningful OOM risk under concurrent load.
• New test TestImageInboundTransformer_TransformRequest_Edit_Multipart_AcceptsImageAboveLegacy4MBLimit validates the relaxed constraint end-to-end.

Confidence Score: 3/5

The limit relaxation itself is correct, but the trace-logging path now base64-encodes up to 800 MB of image data per request, which can exhaust memory under concurrent load.

The core limit change is correct and well-tested. The concern is the trace-logging function that base64-encodes every uploaded image in-process: at 16 x 50 MB a single request can hold ~2.7 GB simultaneously, and concurrent uploads multiply that quickly.

llm/transformer/openai/image_inbound.go — specifically the buildMultipartJSONBody call sites in transformEditRequest and transformVariationRequest.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Incoming HTTP Request] --> B{len body > maxImageBodySize ~816 MB?}
    B -- Yes --> ERR1[Error: request body too large]
    B -- No --> C[Parse multipart parts]
    C --> D{Part has filename?}
    D -- No: field --> E{len value > maxImageFileSize 50 MB?}
    E -- Yes --> ERR2[Error: field too large]
    E -- No --> F[Store as form field]
    D -- Yes: file --> G{imageCount > 16?}
    G -- Yes --> ERR3[Error: too many images]
    G -- No --> H{len data > maxImageFileSize 50 MB?}
    H -- Yes --> ERR4[Error: file too large]
    H -- No --> I{Allowed MIME type?}
    I -- No --> ERR5[Error: unsupported image type]
    I -- Yes --> J[Append to formData.Images]
    J --> C
    F --> C
    C --> K[buildMultipartJSONBody base64-encodes ALL images for trace logging]
    K --> L[Return llm.Request]

_{Reviews (1): Last reviewed commit: "feat: raise image upload limits to 50MB ..." | Re-trigger Greptile}