Show HN: A minimal Flask and Htmx and LangChain (with llama.cpp) interface

A minimal Flask + HTMX + LangChain (with llama.cpp) interface for learning purposes only

❗ This project is a personal learning experiment. It is not production-ready. Do not deploy this without major modifications.

This project was built as a hands-on experiment in combining:

• 🧠 A local LLM backend (via llama.cpp and LangChain)
• ⚡ HTMX for seamless front-end interactivity without JavaScript frameworks
• 🌐 Flask for backend routing and SSE streaming
• 💅 A Neumorphic UI for minimal styling

It’s a prototype, intended for exploring techniques like:

• Streaming LLM output over Server-Sent Events (SSE)
• Managing chat history by session (in SQLite)
• Coordinating LLM calls with a safe in-process queue
• Building interactive web apps with minimal frontend JavaScript using htmx

This project is not suitable for deployment.

• ❌ No authentication or session protection
• ❌ Very basic error handling
• ❌ Blocking, single-threaded queue for LLM calls
• ❌ Input is not sanitized or restricted

It’s meant for educational use only.

• uv
• a compatible GGUF LLM model (e.g. Phi-3.5)
• a relatively fast computer (ideally with a strong GPU)

Download Phi-3.5-mini-instruct-GGUF (tested with the Q5_K_M quantization). Set the CHAT_MODEL_GGUF environment variable to the full path of the .gguf file. Or edit the .env file to include: CHAT_MODEL_GGUF=/path/to/your/model.gguf

Install uv. Then run:

Set FLASK_DEBUG=1 for automatic reloading on code changes.

For CUDA support (Nvidia GPU) you must reinstall the llama-cpp-python library (and have the CUDA toolkit installed):