Polylogue

discussion or speech involving multiple participants, through multiple forms of communication

A Figma-like canvas for visualizing and exploring branching LLM conversations. Each prompt & completion is represented as a tree node, and different models can be selected for each node.

Goals

• Enable multimodal capabilities such as images
• Enable real-time collaborative editing

Backlog

• Allow resizing nodes
• Explore templating conversations for different use cases

Stack

• Next.js & tailwind
• Flask
• Redis
• Firestore
• Langchain
• Together.ai
• GCP

Deployment

This app is deployed using Google App Engine. Deploy backend first.

1. Configure backend/app.yaml and add any required env variables. Reference backend/app.example.yaml.
2. Run gcloud app deploy in backend/.
   • Run gcloud app logs tail -s backend to debug issues.

Deploy frontend second.

1. Configure frontend/app.yaml and add any required env variables. Reference frontend/app.example.yaml.
2. Run NEXT_PUBLIC_BACKEND_ROOT_URL=<APP_ENGINE_BACKEND_URL> yarn build.
   • ex: NEXT_PUBLIC_BACKEND_ROOT_URL=https://backend-dot-polylogue-123456.ew.r.appspot.com yarn build
3. Run gcloud app deploy in frontend/.
   • Run gcloud app logs tail -s frontend to debug issues.