Multi-LLM PDF Chat (Gradio)

Chat with any PDF, querying LLMs individually or let them work out a response among themselves.

Note: this is my experiment in vibe-coding. Don't expect everything to be air-tight.

✨ Features

• Single or multi-model replies, decide per reply – choose OpenAI ChatGPT, Claude Sonnet 4, Gemini 2.5 Pro, or an automatic aggregation of all three.
• One-click controls – five fixed buttons (ChatGPT · Claude · Gemini · ChatGPT + Claude · All)
• Quick Redo - Leave input field empty and click a button to redo the last response
• Aggregator logic – Claude Sonnet 4 coordinates proposer replies and may iterate up to five times before producing a final answer.
• Cost & timeout guardrails – 120 s per request, exponential-back-off retries, and a $5 session budget cap.
• Lean persistence – only user inputs & final replies are stored (Chats/).
• Individual model tabs – separate tabs for ChatGPT, Claude, and Gemini show their last outputs, including proposals that don't appear in the main chat.
• Notification - shows a system notification once the final answer is ready.

🚀 Quick Start

git clone https://github.com/Tamulur/EnsembleChat.git
cd <your-repo>
pip install -r requirements.txt          # or poetry install
export OPENAI_API_KEY=...                # plus ANTHROPIC_API_KEY, GOOGLE_API_KEY
python app.py

Open the Gradio URL, click Select PDF, choose a model button, and start chatting.