RepoMind 🧠🤖

RepoMind is an enterprise-grade, advanced Multi-Agent RAG (Retrieval-Augmented Generation) assistant designed to ingest, analyze, and query complex codebases. Built with state-of-the-art Generative AI frameworks, it combines high-performance vector search, cross-encoder reranking, autonomous multi-agent collaboration, and robust MLOps observability.

---

🚀 Key Features & Tech Stack

• Advanced RAG & Orchestration: Powered by LlamaIndex for intelligent document loading and data structuring.
• Cross-Encoder Reranking: Integrates BAAI/bge-reranker-v2-m3 (top_n = 5) to boost semantic retrieval precision and filter out irrelevant noise.
• Multi-Agent Collaboration: Uses CrewAI to orchestrate specialized agents (Senior Codebase Analyst & Technical Documentation Specialist) working sequentially.
• Vector Database: Uses ChromaDB with persistent disk storage for efficient similarity search.
• Unified LLM Gateway & MLOps:
  • LiteLLM handles model routing and provides a unified interface.
  • Langfuse provides automatic end-to-end tracing and monitoring for every LLM call and agent workflow step.
• Flexible Model Support: Powered by Anthropic Claude Haiku (anthropic/claude-haiku-4-5-20251001) for cloud tasks, alongside local open-weight models via Ollama (gpt-oss:20b and nomic-embed-text for embeddings).
• Containerization: Fully dockerized for seamless deployment.

---

🛠️ Getting Started & Installation

Follow these steps to set up and run RepoMind locally.

1. Prerequisites

• Python 3.12+ installed on your machine (pinned dependency versions are verified against 3.12).
• Ollama installed locally to serve open-source models and embeddings.

2. Clone and Install Dependencies

Clone the repository and install the required Python packages:

git clone https://github.com/mttpzz/repomind.git cd repomind

Create and activate a virtual environment (recommended)

python -m venv venv source venv/bin/activate # On Windows use: venv\Scripts\activate

Install dependencies

pip install --no-cache-dir -r requirements.txt

3. Configure Environment Variables

Copy the example environment file and fill in your API keys:

cp .env.example .env

Open .env and set your configuration:

• ANTHROPIC_API_KEY: Your Anthropic API key (anthropic/claude-haiku-4-5-20251001).
• LANGFUSE_PUBLIC_KEY & LANGFUSE_SECRET_KEY: Your Langfuse tracking keys (optional, but recommended for MLOps).
• Local model names and Ollama endpoint URLs.

4. Start Ollama and Pull Local Models

Make sure your Ollama background service is running:

ollama serve

In a separate terminal window, download the required local embedding model:

ollama pull nomic-embed-text

---

🧪 Running Tests

RepoMind includes unit tests covering the RAG pipeline setup (tests/test_rag.py), the CrewAI agent/LLM wiring (tests/test_crew_manager.py), and the CLI entry point (tests/test_main.py). All external calls (CrewAI, LiteLLM, Ollama embeddings) are mocked, so the suite runs offline — no Anthropic key, Langfuse account, or running Ollama instance required. Run it with pytest:

pytest

Note: these tests only verify that the code wires the right calls with the right parameters — they don't catch integration issues (wrong model name, incompatible library version, misconfigured API key). Always do a real end-to-end ingest + query run (see below) before trusting a change.

---

💻 Usage (CLI Guide)

RepoMind comes with a built-in Command Line Interface (CLI) supporting two primary actions: Ingestion and Querying.

Step A: Ingest a Codebase

Scan, chunk, and index a target source code directory into ChromaDB:

python -m src.main ingest --dir ./path/to/your/target/codebase

Step B: Query the Codebase with Multi-Agent Analysis

Run the multi-agent workflow to analyze the codebase and answer technical prompts:

python -m src.main query --target-dir ./path/to/your/target/codebase --prompt "How is authentication handled in this project?"

---

📊 MLOps & Observability

RepoMind integrates LiteLLM and Langfuse natively. Every query, vector retrieval, reranking step, and multi-agent interaction is automatically logged and tracked. You can inspect token usage, latency, and step-by-step reasoning chains directly on your Langfuse Dashboard (https://cloud.langfuse.com).

---

🐳 Docker Deployment

Build the image:

docker build -t repomind:latest .

The container's entrypoint runs python -m src.main, so the ingest/query subcommand and flags are passed at docker run time. Ollama runs on your host, not inside the container, so point OLLAMA_API_BASE at the host (host.docker.internal on Docker Desktop for Windows/Mac). The target codebase you want to analyze also isn't part of the image — mount it as a volume.

Ingest

docker run --rm --env-file .env
-v "$(pwd)/data:/app/data"
-v "/path/to/target/codebase:/target"
-e OLLAMA_API_BASE=http://host.docker.internal:11434
repomind:latest ingest --dir /target

Query

docker run --rm --env-file .env
-v "$(pwd)/data:/app/data"
-v "/path/to/target/codebase:/target"
-e OLLAMA_API_BASE=http://host.docker.internal:11434
repomind:latest query --target-dir /target --prompt "How is authentication handled in this project?"

• -v $(pwd)/data:/app/data persists the ChromaDB vector store (./data/chroma_db) across ingest and query runs — without it, each container starts with an empty index.
• -v /path/to/target/codebase:/target mounts the codebase you want to ingest/query; --dir/--target-dir must point at the path inside the container (/target), not the host path.
• --env-file .env supplies ANTHROPIC_API_KEY and the optional Langfuse keys — don't bake secrets into the image.

---

📄 License

Distributed under the MIT License. See LICENSE for more information.