ENV_SETUP.md

Environment Variables Setup

This document describes the environment variables needed to run the RAG system.

Required Variables

HF_TOKEN (Required)

Your Hugging Face API token for accessing the Inference API.

How to get it:

1. Go to https://huggingface.co/settings/tokens
2. Click "New token"
3. Give it a name (e.g., "RAG System")
4. Select read permissions
5. Click "Generate"
6. Copy the token (starts with hf_)

Set it:

# Linux/Mac
export HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

# Windows (PowerShell)
$env:HF_TOKEN="hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

# Windows (CMD)
set HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

For Hugging Face Spaces:

• Go to Space Settings → Repository secrets
• Add secret: Name=HF_TOKEN, Value=your token

Optional Variables

HF_MODEL

The Hugging Face model to use for text generation.

Default: meta-llama/Llama-3.2-3B-Instruct

Other options:

• mistralai/Mistral-7B-Instruct-v0.2
• meta-llama/Meta-Llama-3-8B-Instruct

export HF_MODEL=meta-llama/Llama-3.2-3B-Instruct

EMBEDDING_MODEL

The sentence transformer model for embeddings (runs locally).

Default: all-MiniLM-L6-v2

export EMBEDDING_MODEL=all-MiniLM-L6-v2

CHROMA_PERSIST_DIR

Directory for ChromaDB vector database storage.

Default: ./chroma_db

export CHROMA_PERSIST_DIR=./chroma_db

DOCUMENTS_DIR

Directory containing source documents.

Default: ./documents

export DOCUMENTS_DIR=./documents

PROCESSED_DOCS_DIR

Directory for processed markdown documents.

Default: ./processed_docs

export PROCESSED_DOCS_DIR=./processed_docs

DEFAULT_N_RESULTS

Number of context chunks to retrieve for each query.

Default: 5

export DEFAULT_N_RESULTS=5

SIMILARITY_THRESHOLD

Minimum similarity score for retrieved chunks.

Default: 0.5

export SIMILARITY_THRESHOLD=0.5

CHUNK_SIZE

Size of text chunks for processing.

Default: 1000

export CHUNK_SIZE=1000

CHUNK_OVERLAP

Overlap between consecutive chunks.

Default: 200

export CHUNK_OVERLAP=200

GRADIO_SHARE

Whether to create a public Gradio share link.

Default: False (set to True for 72-hour public link)

export GRADIO_SHARE=True

GRADIO_SERVER_PORT

Port for the Gradio server.

Default: 7860

export GRADIO_SERVER_PORT=7860

Complete Example

Create a .env file in the project root:

# Required
HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

# Optional (uncomment to override defaults)
# HF_MODEL=meta-llama/Llama-3.2-3B-Instruct
# EMBEDDING_MODEL=all-MiniLM-L6-v2
# CHROMA_PERSIST_DIR=./chroma_db
# DOCUMENTS_DIR=./documents
# PROCESSED_DOCS_DIR=./processed_docs
# DEFAULT_N_RESULTS=5
# SIMILARITY_THRESHOLD=0.5
# CHUNK_SIZE=1000
# CHUNK_OVERLAP=200
# GRADIO_SHARE=False
# GRADIO_SERVER_PORT=7860

Then load it before running:

# Using python-dotenv (recommended)
pip install python-dotenv
# Add to your script: from dotenv import load_dotenv; load_dotenv()

# Or manually source it
source .env  # Linux/Mac

Verification

To verify your environment is set up correctly:

python -c "import os; print('HF_TOKEN:', 'SET' if os.getenv('HF_TOKEN') else 'NOT SET')"

Should output: HF_TOKEN: SET