Ask questions in plain English about any file in your project:
PDFs, Word, Excel, Python, JavaScript, Markdown, logs… everything is instantly searchable and remembered forever.
Built with FastAPI + Groq (Llama 3.1) + LangChain 1.1 + Chroma + PostgreSQL
Fully Dockerized — runs anywhere in seconds.
| Feature | Status | Description |
|---|---|---|
| Chat with any document/code | Done | PDF, DOCX, XLSX, TXT, .py, .js, .md, .log … |
| Persistent conversation memory | Done | Postgres-backed — survives restarts |
| Drag & drop file upload → auto-indexing | Done | No terminal needed |
| Real-time sources shown | Done | See exactly which file(s) the answer came from |
| Zero-config Docker setup | Done | docker compose up and you’re live |
| Model caching | Done | First start ~20 s, then < 3 s startup forever |
| Works offline (after first download) | Done | Only Groq API key needed |
| Your Laptop / PC | Approx. number of average books* you can load | Startup time (cold) | Query speed | Notes |
|---|---|---|---|---|
| 8–12 GB RAM | 50 – 150 books | 30–60 sec | 2–4 sec | Starts swapping |
| 16 GB RAM (most laptops) | 250 – 450 books | 15–30 sec | 1–2 sec | Sweet spot for most users |
| 32 GB RAM | 600 – 1 000 books | 10–20 sec | < 1.5 sec | Perfect for power users |
| 64 GB+ RAM (M2/M3 Max, desktop) | 1 200+ books | 8–12 sec | < 1 sec | Buttery smooth |
*One “average book” ≈ 300–500 pages ≈ 5–50 MB PDF/text
| 500 books (mixed PDF + text) | ~15–60 GB total (raw files + Chroma vector DB) |
|---|---|
| HuggingFace embedding model | ~350 MB (cached once) |
→ The real limits are RAM and disk space, not the code.
Everything runs completely locally — no data ever leaves your machine.
Switch the vector store from local Chroma to Pinecone / Qdrant / Weaviate (one-line change) and scale to tens of thousands of documents with zero slowdown.
# 1. Clone
git clone https://github.com/yourname/Codebase-GPT.git
cd Codebase-GPT
# 2. Put your Groq key in .env, create your own at https://groq.com/
echo "GROQ_API_KEY=gsk_your_key_here" > .env
# 3. Start everything
docker compose up --buildOpen → http://127.0.0.1:8000
Drop files → ask anything → enjoy!
Codebase-GPT/
├── app/
│ ├── templates/index.html # beautiful chat UI
│ ├── main.py # FastAPI + RAG logic
│ ├── ingest.py # File indexing script
│ └── requirements.txt
├── repo-to-index/ # ← DROP YOUR FILES HERE
├── data/chroma_db/ # ← vector database (auto-created)
├── models/ # ← cached HuggingFace models
├── Dockerfile
├── docker-compose.yml
└── .env # your Groq key
| Layer | Technology |
|---|---|
| Web Framework | FastAPI + Uvicorn |
| LLM | Groq (Llama 3.1 8B or 70B) |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Vector Store | Chroma (langchain-chroma) |
| Document Loaders | Unstructured (PDF/DOCX/Excel) + TextLoader |
| Memory | PostgreSQL + SQLChatMessageHistory |
| Frontend | Pure HTML + jQuery (no React/Vue needed) |
- Speed: Groq is the fastest inference provider in 2025
- Cost: Free tier is generous, paid is cheap
- Privacy: All your files stay on your machine / server
- Simplicity: Zero JavaScript framework → easy to customize
Just use the upload area at the top of the web page.
# Put files here
cp your-file.pdf repo-to-index/
# Re-index (takes seconds)
docker exec -it <container_name> python app/ingest.py
# or locally:
python app/ingest.pyWe love contributions! Here’s how to help:
- Fork the repo
- Create your feature branch (
git checkout -b feature/amazing-idea) - Commit (
git commit -m 'Add streaming responses') - Push and open a Pull Request
- Streaming token-by-token responses
- Dark mode toggle
- Chat export (JSON/Markdown)
- Authentication (OAuth2 / simple password)
- Support for images in PDFs (OCR)
- Better Excel table rendering
All contributions are welcome — even tiny fixes!
MIT License — do whatever you want with it.
Made with love by developers who were tired of Ctrl+F-ing through 50 files every day.
Enjoy your private, lightning-fast, always-remembering document assistant!