Much better handling of tool call outputs!

In this version:

Critical bugfix:

• All tool calls are once again correctly streamed to the frontend chat client

New features:

1. Setup page now allows customization of all salient assistant features, including:

• System prompt
• Tool selection
• File upload for file search tool

2. In the chat client, tool calls are prefixed with a helpful message indicating what kind of tool call it is (so file search tool calls are no longer just an empty container)
3. All code interpreter tool call outputs are now neatly served to the user:

• Printed code outputs are formatted like "> Hello world!" to distinguish them from tool call deltas
• File outputs are made user-downloadable through a link to a dedicated FastAPI endpoint for that purpose
• Images are correctly embedded and displayed in the chat interface

5. Files uploaded for use with the file search tool are saved in a local application folder and linked from file citation annotations for user download (since OpenAI does not store or provide for download of these files from its server)

Starter template to create and chat with an OpenAI assistant with Python + HTMX!

In this release:

1. Create your assistant on the setup page! Select your assistant's model and tools, and upload documents to provide the assistant a custom vector store for file search.
2. Chat with your assistant on the chat page! The application streams assistant tool call and assistant message text deltas via SSE to the frontend, where the HTMX SSE extension receives them and hands them off to the marked Javascript library for conversion from markdown to HTML before swapping them in.
3. Assistant can use the code interpreter, the file search tool, or any custom Python function you write (see utils/custom_functions.py for an example get_weather function).

Excited to see what you build on this foundation!