vorp 0.1.3

A terminal-based AI companion.

🧠 vorp

vorp is a terminal-based AI pair programmer. It indexes your codebase, allowing you to ask context-aware questions and retrieve relevant code snippets without leaving your command line environment.

Note: This project is under active development.

🚀 Key Features

• RAG (Chat with Codebase): Index any project folder to enable context-aware queries.
  • Project Isolation: Uses a global vector database with metadata filtering. Context from Project A will never leak into Project B.
  • Local Storage: All embeddings are stored locally in ~/.vorp_rag_db.
• Multi-Model Support: Integrates with Groq and Google Gemini to provide access to models like Llama 3.3, DeepSeek R1, and Gemini 2.5 Pro.
• Session Persistence: Chat history is saved locally, allowing you to resume sessions later.
• Context Management: Manually inject specific files into the context window for targeted assistance.
• Cross-Platform: Designed to work consistently on Windows, macOS, and Linux.

🛠️ Installation

Prerequisites

• Python 3.10+
• Git

Steps

1. Clone the repository:

   git clone https://github.com/SiddharthBayapureddy/vorp.git
   cd vorp
   

   (Note: If the repository is renamed to vorp, clone that instead.)

2. Create a virtual environment:

   • Windows:

     python -m venv venv
     .\venv\Scripts\activate
     

   • macOS/Linux:

     python3 -m venv venv
     source venv/bin/activate
     

3. Install dependencies:

   pip install -e .
   

4. Configure API Keys: Create a .env file in the root directory and add your keys:

   GROQ_API_KEY=your_key_here
   GEMINI_API_KEY=your_key_here
   

🎮 Usage

Start the application:

vorp

Interactive Commands

Command	Description
/index <path>	Scans and indexes the specified directory. This creates a searchable vector index for RAG.
/rag	Toggles RAG mode on or off. When enabled, the assistant retrieves context from the indexed project.
/add <file>	Loads the content of a specific file into the active chat context.
/context	Displays a list of currently loaded files and the active RAG project path.
/clear	Clears the terminal screen.
/exit-v	Exits the application and saves the current chat history.
/exit	Exits the application and deletes the current chat history.

CLI Arguments

You can configure vorp at startup using these flags:

Flag	Description
--model <id>	Starts the session with a specific model (e.g., groq/llama-3.3-70b-versatile).
--list	Lists all supported models and their IDs, then exits.
--help	Displays the help message.

Example:

vorp --model "gemini/gemini-2.5-pro"

🏗️ Architecture

The Retrieval-Augmented Generation (RAG) system in vorp is built for speed and privacy. Here is how it works under the hood:

1. Ingestion & Chunking:

   • When you run /index, the system walks through your project directory.
   • Files are read and split into smaller segments using a Sliding Window approach (1000 characters with 200 character overlap). This ensures that context at the boundaries of chunks is preserved.

2. Embedding Generation:

   • Each chunk is passed through the all-MiniLM-L6-v2 model. This is a lightweight, high-performance model that runs locally on your CPU.
   • The model converts the text code into a 384-dimensional vector (a list of numbers representing the semantic meaning).

3. Vector Storage (ChromaDB):

   • These vectors are stored in ChromaDB, a persistent local vector database located at ~/.vorp_rag_db.
   • Isolation Layer: Every vector is tagged with a project_id metadata field (the absolute path of the project). This acts as a strict filter, ensuring that queries only search within the active project's scope.

4. Retrieval (Cosine Similarity):

   • When you ask a question in RAG mode, your query is embedded using the same model.
   • The database performs a similarity search (using Cosine Similarity) to find the top 5 chunks that are mathematically closest to your query.
   • This retrieval is strictly filtered by the active project_id.

5. Context Injection:

   • The retrieved code snippets are formatted and injected into the LLM's system prompt.
   • The LLM then generates an answer using this retrieved knowledge, allowing it to "see" your code.

🔮 Roadmap

• File Editing: Capabilities for the agent to autonomously modify files.
• Command Execution: Safe execution of shell commands for testing and linting.
• Diff View: Enhanced visualization of code changes.

🤝 Contributing

Contributions are welcome. Please open an issue or submit a pull request for any improvements.