ChatGPT on your hard drive, stripped to the essentials

What it does This is a bare-bones demo for querying your own documents through ChatGPT. It uses langchain and chromadb to ground OpenAI API prompts in local text or PDF files, so a question like “what is my dog’s name” is answered from data/data.txt rather than the model’s training data.

The interesting bit The project doesn’t pretend to be a framework; it is openly glue code meant to accompany a YouTube tutorial. That transparency makes it a decent reference implementation for developers who want to see how document retrieval is wired to an LLM call without hidden abstraction layers.

Key highlights

• Built on langchain, openai, chromadb, tiktoken, and unstructured.
• Reads local text and PDFs from a data/ directory.
• Requires an OpenAI API key; no local LLM inference.
• Centered around a minimal chatgpt.py script with a simple constants file.
• 1,699 stars suggest the tutorial audience found it a useful starting point.

Caveats

• The README is extremely terse and omits details on chunking strategy, embedding models, or how unstructured parses complex PDFs.
• Path and API-key handling look like tutorial-grade conventions, not production patterns.
• No discussion of API cost, rate limits, or data privacy beyond the local vector store.

Verdict Worth a look if you want a dead-simple, no-magic reference for wiring OpenAI to local documents. Look elsewhere if you need a configurable CLI, an offline stack, or production-grade ingestion pipelines.