PDF Support in the Playground

[mmabrouk]

Problem Statement

The playground currently supports text and images, but many LLM providers now support PDF files. Users working with document analysis, extraction, or question answering on PDFs cannot test these workflows in the playground.

Proposed Solution

We are adding PDF support to the playground for models that support it (OpenAI, Gemini, etc.). Users will be able to provide PDFs in three ways:

• Base64 encoded content
• Public URLs pointing to PDF files
• File IDs from provider-specific APIs (e.g., OpenAI Files API)

The playground will handle the appropriate format conversion based on the selected model provider. This support will extend to human evaluation as well, allowing evaluators to review model responses on PDF inputs.

[ashrafchowdury]

Thank you for opening the discussion, @mmabrouk. PDF support has been implemented for supported providers and is now available in the playground.