Feature Request: Preserve original filename when downloading media attachments

[YouJun-IWON]

Describe the feature or potential improvement

Feature Description

When uploading files using LangfuseMedia, the original filename is lost during S3 storage, causing downloads to use the media ID with a .bin extension instead of the original filename.

Current Behavior

Upload:
from langfuse import get_client
from langfuse.media import LangfuseMedia

langfuse_client = get_client()

Upload a PowerPoint file

with open("presentation.pptx", "rb") as f:
file_bytes = f.read()

pptx_media = LangfuseMedia(
content_bytes=file_bytes,
content_type="application/vnd.openxmlformats-officedocument.presentationml.presentation"
)

langfuse_client.update_current_trace(
input={
"presentation.pptx": pptx_media,
},
metadata={
"file_info": {
"filename": "presentation.pptx",
"size_mb": 2.5
}
}
)Result:

• ✅ Langfuse UI displays the file in the input section
• ✅ Metadata shows the original filename
• ❌ Download saves as WY7P2DC7gRxnrTRg0dp3Be.bin instead of presentation.pptx

Expected Behavior

When downloading a media attachment from the Langfuse UI, the file should be downloaded with:

1. Original filename: presentation.pptx
2. Correct extension: .pptx (not .bin)

Proposed Solution

Add a filename parameter to the LangfuseMedia class:

LangfuseMedia(
content_bytes=file_bytes,
content_type="application/pdf",
filename="document.pdf" # ✨ New parameter
)This filename should be:

1. Stored alongside the media metadata in S3
2. Used in the Content-Disposition header when serving download links
3. Displayed in the Langfuse UI

Use Cases

This feature is critical for:

• 📊 Document analysis workflows: Users need to identify which document was analyzed
• 🔍 Audit trails: Original filenames are important for compliance
• 👥 Team collaboration: Team members need to recognize files by their original names
• 📁 Multi-file traces: Distinguishing between multiple files in a single trace

Current Workaround

We currently store the filename in metadata, but this requires users to:

1. Check metadata to find the original filename
2. Manually rename the downloaded .bin file
3. This is error-prone and reduces user experience

Additional Context

• Similar issue: Manually creating file names for images saved in Minio #5898 (Manually creating file names for images in MinIO)
• Related discussion: Support for more modalities and base64 content #3004 (Multi-modal support feedback)
• SDK Version: Python SDK v3
• Affected file types: All non-image/audio files (PPTX, DOCX, XLSX, HWPX, etc.)

Benefits

• ✅ Improved user experience
• ✅ Better file management
• ✅ Reduced confusion
• ✅ Professional presentation of traces
• ✅ Maintains file context across workflow

Are you willing to contribute?

Yes, I can contribute a PR if provided with guidance on:

• Where to add the filename parameter in the SDK
• How to store it in the media metadata
• How to modify the download endpoint to use it

---

Thank you for considering this feature request! 🙏

Additional information

No response

[hassiebp]

Thanks for raising this! Could you please retry your PPTX upload with Langfuse Cloud? We have recently shipped improvements to media handling incl PPTX that should lead to a correct file extension (while the file name is still the media ID)