Docs-Reference: A Common Lisp RAG System for Local Models

2 pointsposted 10 hours ago
by skarnati20

1 Comments

skarnati20

10 hours ago

First post in this community, and I wanted to share my first completed project in Common Lisp.

This is a very basic Local LLM Chat + RAG implementation, but it has some features I couldn't find in the mainstream products.

I built this because I found Claude/GPT overkill for lots of programming questions I had, and I needed something that could work offline. It indexes a URL you provide and lets you ask questions to a local LLM (Ollama) with that information.

Some Features * Multiple Retrieval Methods — (BM25, Dense Vector Search, ColBERT). You can choose whichever combination of these you want. * Agentic RAG — An Agent determines when there is enough context to answer a question. You can turn it off if you want speed. * Fully Local Chat — Uses Ollama so you can run any model you want, so long as you serve it locally. * Asynchronous Chat — Allows you to ask multiple questions at once which helps with multi-tasking and getting around local models' slow completion times.

Acknowledgement: This project was an extension/combination of some projects in Mark Watson's Loving Common Lisp, or the Savvy Programmer's Secret Weapon. Definitely check it out since it gave me a lot of ideas for fun side projects.