This repo is a minimal Retrieval-Augmented Generation (RAG) system that combines both:

• Short-term memory using Redis (key-value)
• Long-term semantic memory using FAISS (vector search)
• Embeddings generated locally using Mistral via Ollama

• Python CLI interface for chatting with LLMs
• Memory summaries of user+assistant exchanges stored in Redis
• User inputs and summarized replies embedded and stored in FAISS
• Vector search for semantically relevant recall (even after a restart)
• Modular design via gpt_abstraction.py and vector_store.py
• All local, no external APIs required

• Docker
• Ollama installed on your host (for running Mistral)
• Python 3.11+ if you want to run parts outside the container

git clone https://github.com/tdenton8772/chat_cli_vector
cd chat_cli_vector

docker compose build

This sets up:

• A Python container for the CLI
• Redis for memory storage
• Ollama for embedding and model inference
• Persistent volumes for Redis and model data

You need to pull mistral manually before running the CLI:

ollama pull mistral

This stores the model in a volume accessible to the containerized Ollama service.

./run.sh

You'll be prompted to select a model and either start a new conversation or continue an existing one.

/list             Show all saved conversations
/switch <id>      Switch to another conversation
/delete <id>      Delete a conversation
/recap            View Redis-stored memory summaries
/vectors          View all stored FAISS memory entries
/help             Show this help menu
exit / quit       End the session

.
├── chat_cli.py            # Main CLI loop
├── hybrid_memory.py       # Combines Redis + FAISS memory
├── vector_store.py        # FAISS wrapper with persistence
├── embedding.py           # Calls Mistral via Ollama for embeddings
├── gpt_abstraction.py     # Switches between Mistral / OpenAI / Claude
├── memory_summarizer.py   # NLP summarizer (used before storing in Redis)
├── config.py              # Model names, embedding dim, top_k, etc.
├── run.sh                 # Host launch script
├── entrypoint.sh          # CLI startup
├── docker-compose.yml
├── Dockerfile

Notes on Model Support Fully tested: Mistral via Ollama

Claude + OpenAI supported via gpt_abstraction.py, but not tested in this setup

Easily swappable with your own backend model service

PRs and feedback welcome!