Skip to main content

​
Quick start

This guide will walk you through setting up API keys, loading your first repository, and asking questions.
Before you begin: Make sure you’ve completed the installation steps.

​
Step 1: Get your API keys

RepoRAGX requires two API keys to function:
1

GitHub personal access token

  1. Go to github.com/settings/tokens
  2. Click “Generate new token” → “Generate new token (classic)”
  3. Give it a descriptive name like “RepoRAGX”
  4. Select the repo scope (or just public_repo if you only need public repositories)
  5. Click “Generate token” and copy the token immediately
GitHub only shows the token once. Save it in a secure location.
2

Groq API key

  1. Go to console.groq.com
  2. Sign up for a free account (no credit card required)
  3. Navigate to console.groq.com/keys
  4. Click “Create API Key”
  5. Give it a name and copy the API key
Groq’s free tier includes generous rate limits for the llama-3.3-70b-versatile model.

​
Step 2: Run RepoRAGX

With your virtual environment activated, start RepoRAGX: 
python -m src.main
You’ll see the RepoRAGX banner and be prompted for credentials: 
/**
 *    __________                    __________    _____    ____________  ___
 *    \______   \ ____ ______   ____\______   \  /  _  \  /  _____/\   \/  /
 *     |       _// __ \\____ \ /  _ \|       _/ /  /_\  \/   \  ___ \     / 
 *     |    |   \  ___/|  |_> >  <_> )    |   \/    |    \    \_\  \/     \ 
 *     |____|_  /\___  >   __/ \____/|____|_  /\____|__  /\______  /___/\  \
 *            \/     \/|__|                 \/         \/        \/      \_/
 */

Chat with your github repository

GitHub Personal Access Token:

​
Step 3: Configure your session

Enter your credentials and repository details when prompted: 
GitHub Personal Access Token: ghp_xxxxxxxxxxxxxxxxxxxx
Groq API Key: gsk_xxxxxxxxxxxxxxxxxxxx
Model Name (default: llama-3.3-70b-versatile): [press Enter]
Repo (owner/repo): AnmolTutejaGitHub/RepoRAGX
Branch (default: main): main
You can use any model supported by Groq. Popular options:
  • llama-3.3-70b-versatile (default, best quality)
  • llama-3.1-70b-versatile (alternative)
  • mixtral-8x7b-32768 (faster, good for smaller queries)
See the full list at console.groq.com/docs/models.
Use the format owner/repo:
  • ✅ facebook/react
  • ✅ microsoft/vscode
  • ✅ AnmolTutejaGitHub/RepoRAGX
  • ❌ https://github.com/facebook/react (don’t include the full URL)
The default branch is main. You can specify any branch name:
  • main
  • master
  • develop
  • feature/new-auth

​
Step 4: Loading the repository

RepoRAGX will now fetch and process the repository: 
Initilizing github loader.....
Fetching files from github....
Loaded 42 files from github!
Splitting documents into chunks...
chunking completed
handling document embedding generation using sentence transformer...
Loading embedding model: all-MiniLM-L6-v2
Model loaded successfully. Embedding dimension: 384
Generating embeddings for 187 texts...
Batches: 100%|████████████████████| 6/6 [00:03<00:00,  1.89it/s]
Generated embeddings with shape: (187, 384)
Initilizing vector database....
Vector store initialized. Collection: AnmolTutejaGitHub_RepoRAGX
Existing documents in collection: 0
Adding 187 documents to vector store...
Successfully added 187 documents to vector store
Total documents in collection: 187
Initilizing RAG Retriver pipeline
Initializing Groq LLM...
First run: The embedding model (all-MiniLM-L6-v2) will be downloaded automatically (~90MB). This only happens once.

​
What’s happening behind the scenes

1

File loading

GitHubCodeBaseLoader fetches all files from the repository, filtering out:
  • Binary files (.png, .jpg, .exe, etc.)
  • Dependencies (node_modules/, venv/, .git/)
  • Generated files (.pyc, .class, .min.js)
See the full exclusion list in src/rag/github_codebase_loader.py:3-24
2

Text chunking

TextSplitter uses language-aware splitting with RecursiveCharacterTextSplitter:
  • Chunk size: 1000 characters
  • Chunk overlap: 200 characters
  • Supports 25+ languages (Python, JavaScript, Java, Go, Rust, etc.)
Language detection happens automatically based on file extension. See src/rag/text_splitter.py:3-50
3

Embedding generation

EmbeddingManager generates 384-dimensional vectors using Sentence Transformers:
# From src/rag/embedding_manager.py:5
model_name = "all-MiniLM-L6-v2"
Each code chunk is converted to a vector for similarity search.
4

Vector storage

VectorStore persists embeddings in ChromaDB:
  • Location: ~/.RepoRAGX/vector_store/
  • Distance metric: Cosine similarity
  • Collection name: owner_repo (e.g., facebook_react)
See src/rag/vector_store.py:13-40

​
Step 5: Ask questions

Once loading is complete, you can start asking questions: 
Ask anything ('exit' to quit): How does the text splitter work?
RepoRAGX will:
  1. Convert your query to a vector embedding
  2. Search ChromaDB for the top 5 most similar code chunks
  3. Send the retrieved context to Groq’s LLM
  4. Return a context-aware answer

​
Example queries

Ask anything ('exit' to quit): What is the project structure?
Ask anything ('exit' to quit): How does the RAG pipeline work?
Ask anything ('exit' to quit): What libraries are used for embeddings?

​
Sample output

Ask anything ('exit' to quit): How are embeddings generated?

Running RAG for query: How are embeddings generated?
Retrieving documents for query: 'How are embeddings generated?'
Top K: 5, Score threshold: 0.0
Generating embeddings for 1 texts...
Retrieved 5 documents (after filtering)

Embeddings are generated using the Sentence Transformers library with the 
all-MiniLM-L6-v2 model. The EmbeddingManager class in src/rag/embedding_manager.py 
handles this process:

1. The model is loaded in the _load_model() method (line 11)
2. The generate_embeddings() method takes a list of texts
3. It uses model.encode() to convert texts to 384-dimensional vectors
4. Progress is shown with a progress bar via show_progress_bar=True
5. The method returns a numpy array of shape (num_texts, 384)

The same model is used for both document embeddings and query embeddings to ensure 
compatibility during similarity search.

Ask anything ('exit' to quit): exit

​
Understanding retrieval

When you ask a question, RepoRAGX performs similarity search: 
def retrieve(self, query, top_k=5, score_threshold=0.0):
    # 1. Convert query to embedding
    query_embedding = self.embedding_manager.generate_embeddings([query])[0]
    
    # 2. Query ChromaDB
    results = self.vector_store.collection.query(
        query_embeddings=[query_embedding.tolist()],
        n_results=top_k
    )
    
    # 3. Calculate similarity scores
    for i, (doc_id, document, metadata, distance) in enumerate(...):
        similarity_score = 1 - distance  # Convert distance to similarity
        
        if similarity_score >= score_threshold:
            retrieved_docs.append({
                'content': document,
                'metadata': metadata,
                'similarity_score': similarity_score,
                'rank': i + 1
            })
    
    return retrieved_docs
Pro tip: The more specific your question, the better the results. Instead of “How does this work?”, try “How does the EmbeddingManager generate vector embeddings?”

​
Advanced configuration

​
Adjusting retrieval parameters

You can modify retrieval settings in src/rag/rag_retriever.py:7: 
def retrieve(self, query, top_k=5, score_threshold=0.0):
  • top_k: Number of chunks to retrieve (default: 5)
  • score_threshold: Minimum similarity score (0.0 = no filtering)

​
Changing chunk size

Modify text splitting in src/rag/text_splitter.py:53: 
def __init__(self, documents, chunk_size=1000, chunk_overlap=200):
  • chunk_size: Maximum characters per chunk
  • chunk_overlap: Overlapping characters between chunks

​
Using different models

Groq supports multiple models. Try these alternatives: 
Model Name (default: llama-3.3-70b-versatile): mixtral-8x7b-32768
See src/rag/groq_llm.py:9-26 for LLM configuration.

​
Persistent storage

RepoRAGX stores vector embeddings locally: 
# View stored repositories
ls ~/.RepoRAGX/vector_store/

# Each repository gets its own ChromaDB collection
# Named as: owner_repo (e.g., facebook_react)
Vector stores are recreated each time you load a repository. If you want persistent storage, comment out lines 19-23 in src/rag/vector_store.py:
# try:
#     self.client.delete_collection(name=self.collection_name)
#     print(f"Deleted existing collection: {self.collection_name}")
# except Exception:
#     pass

​
Troubleshooting

Error: Bad credentials
Your GitHub token is invalid or expired. Generate a new one at github.com/settings/tokens.
Error: API rate limit exceeded
GitHub: Wait for the rate limit to reset, or use a token with higher limits. Groq: Free tier has rate limits. Wait a moment between queries.
No relevant context found to answer the question.
Try:
  1. Rephrasing your question to be more specific
  2. Using keywords that appear in the codebase
  3. Lowering score_threshold in src/rag/rag_retriever.py:7
Error loading model all-MiniLM-L6-v2
Check your internet connection. The model downloads from Hugging Face (~90MB). If it continues to fail, try:
pip install --upgrade sentence-transformers

​
Exit the session

To exit RepoRAGX, type exit at the prompt: 
Ask anything ('exit' to quit): exit

​
Next steps

How it works

Deep dive into RepoRAGX’s internals

API reference

Explore the Python API

Configuration

Customize API keys and models

Examples

See real-world usage examples

Build docs developers (and LLMs) love

Get started for freeTalk to us