An intelligent agent framework with integrated RAG capabilities, built on Gemini CLI architecture. The system combines a general-purpose agent engine with specialized RAG tools for document indexing, semantic search, and result reranking.

Core Design: Universal Agent Framework + Professional RAG Toolchain

• Agent Engine: Based on proven Gemini CLI architecture with turn-based conversation management, tool scheduling, and streaming responses
• RAG Pipeline: Document processing → Vector embedding → Semantic search → Cross-encoder reranking
• Multi-Model Support: Gemini, Claude, OpenAI for LLM; Gemini embedding-001 for vectorization; Qwen3-Reranker-8B for reranking

Intelligent Agent Capabilities

• Autonomous decision-making and workflow orchestration
• Context-aware tool selection and parameter optimization
• Dual history mechanism with intelligent compression
• Real-time streaming responses with tool execution feedback

Advanced RAG Implementation

• Multi-format document processing (PDF, Office, Markdown, code, images with OCR)
• Configurable chunking strategies (paragraph, fixed-size, sentence, auto)
• Semantic search with metadata filtering and threshold controls
• Cross-encoder reranking for precision improvement

Production-Ready Infrastructure

• Asynchronous architecture with concurrent tool execution
• Comprehensive monitoring with OpenTelemetry integration
• Multi-environment configuration management
• Graceful error handling and fallback mechanisms

├── packages/
│   ├── core/          # Agent engine and API services
│   ├── cli/           # Command-line interface
│   └── web/           # React web interface (MVP)
├── project_tools/     # RAG-specific tools
│   ├── doc_index_tool.py      # Document indexing
│   ├── vector_search_tool.py  # Semantic search
│   ├── rerank_tool.py         # Result reranking
│   └── collection_info_tool.py # Database management
└── test_documents/    # Sample AI/ML knowledge base

Prerequisites

• Python 3.9+
• Qdrant vector database (local or cloud)
• GOOGLE_API_KEY or GEMINI_API_KEY for embeddings and LLM
• Optional: ANTHROPIC_API_KEY for Claude models
• Optional: OPENAI_API_KEY for OpenAI models
• Optional: Qwen3-Reranker-8B model for reranking

Basic Setup

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env with your API keys and database settings

# Start Qdrant (if local)
docker run -p 6333:6333 qdrant/qdrant

# Run CLI interface
cd packages/cli && python cli.py

Complexity Notice: This project requires significant infrastructure setup including local vector database deployment and large language model hosting. It is designed as an architectural reference rather than a plug-and-play solution.

Resource Requirements:

• Qdrant vector database (2GB+ RAM recommended)
• Qwen3-Reranker-8B model (12GB VRAM for GPU inference)
• Stable internet connection for Gemini API calls

Recommended Use Cases:

• Learning advanced RAG implementation patterns
• Understanding agent-driven architecture design
• Adapting components for custom RAG solutions
• Research and development reference

MIT License - See LICENSE file for details.