Claude Code Instructions

This document provides instructions for Claude Code when working on this repository.

Project Overview

This is a multi-model AI chatbot built with Flask. The project has two parallel versions:

1. Original files - Full-featured version supporting multiple LLM providers
2. New files (*_new.*) - Trimmed version supporting only OpenAI o3 and o1-pro models

File Structure

Original Files (DO NOT MODIFY)

File	Purpose
app.py	Original backend with all models (GPT-4o, GPT-4.1, o3, o4-mini, o3-mini, o1-pro, Claude, Gemini)
templates/index.html	Original frontend with full model selection
templates/index2.html	Original file viewer

New Files (Active Development)

File	Purpose
app_new.py	Trimmed backend - only o3 and o1-pro models
templates/index_new.html	Trimmed frontend - only o3 and o1-pro in dropdown
templates/index2_new.html	File viewer for new version

Critical Rules

1. Preserve Original Files

NEVER modify the following files:

• app.py
• templates/index.html
• templates/index2.html

These files serve as reference implementations. All new development should occur in the *_new.* files.

2. When Making Changes

• All backend changes go in app_new.py
• All frontend changes go in templates/index_new.html or templates/index2_new.html
• If you need to reference original implementation, read from app.py or templates/index.html

3. Running the Application

To run the new trimmed version:

python app_new.py

To run the original full version:

python app.py

Current State of New Files

app_new.py

• Supported models: o3, o1-pro only
• API: OpenAI Responses API (non-streaming)
• Both models use: store=True, reasoning={"effort": "high"}
• Removed: Anthropic client, Gemini client, Claude thinking flag
• Default model: o3

templates/index_new.html

• Model dropdown: Only o3 and o1-pro options
• Removed: Claude Extended Thinking checkbox
• Streaming: Hardcoded to false (both models use non-streaming)

Adding New Features

When adding new functionality:

1. First check if similar functionality exists in the original files
2. Implement changes only in the *_new.* files
3. Test thoroughly before considering backporting to original files
4. Document any new endpoints or features in this file

API Endpoints (New Version)

Endpoint	Method	Description
/	GET	Main chat interface
/get-token	GET	Generate unique session token
/stream	GET	Process prompt and return response
/select_model	POST	Change active model
/save	POST	Save conversation
/autosave	POST	Auto-save conversation
/load	GET	Load previous conversation
/reset	POST	Reset conversation
/shutdown	POST	Shutdown server
/filecontents	GET	File viewer interface
/open_folder	POST	Open folder picker
/load_selection	POST	Load selected files
/update-math-render	POST	Toggle math rendering

Notes

• The new version removes all streaming logic since both o3 and o1-pro use the OpenAI Responses API which returns complete responses
• Session management and conversation persistence work identically to the original
• File viewer functionality is preserved unchanged