<!– generated-by: gsd-doc-writer –> Embenx Documentation (v1.5.1) 🚀 ==============================

The Agentic Memory Layer & Universal Retrieval Toolkit.

Embenx is a high-performance Python library designed for the modern AI ecosystem. It bridges the gap between raw vector indices and full-blown vector databases, providing a unified API for 20+ vector backends, advanced filtering, hybrid search, and specialized memory structures for AI agents.

Contents:

Key Capabilities

  • Unified Collection API: A table-like interface for managing vectors and metadata seamlessly.

  • Hybrid Search (RRF): Combine dense semantic search with sparse BM25 keyword search using Reciprocal Rank Fusion.

  • Agentic Memory (MCP): Built-in Model Context Protocol server for instant integration with Claude, GPT-5, and autonomous agents.

  • Advanced Visuals: Built-in 3D HNSW Graph Visualizer, RAG Playground, and Embenx Explorer.

  • Research-Driven Optimizations: Implementation of state-of-the-art algorithms like ClusterKV, TurboQuant (1-bit quantization), and Echo (Temporal Memory).

  • 20+ Vector Backends: Unified support for FAISS, ScaNN, USearch, HNSWLib, Annoy, pgvector, LanceDB, Milvus, Qdrant, Chroma, Weaviate, DuckDB, Vespa, Elasticsearch, OpenSearch, and BM25.

  • Multimodal Support: Native handling of image embeddings (CLIP) and cross-modal retrieval.

Quick Start

Install Embenx:

pip install embenx

Simple semantic search:

from embenx import Collection
import numpy as np

# Initialize a collection with FAISS-HNSW
col = Collection(dimension=768, indexer_type="faiss-hnsw")

# Add data with metadata
vectors = np.random.rand(100, 768).astype('float32')
metadata = [{"id": i, "text": f"Document {i}", "category": "news"} for i in range(100)]
col.add(vectors, metadata)

# Search with metadata filtering
results = col.search(query_vector, top_k=5, where={"category": "news"})

for meta, dist in results:
    print(f"Found: {meta['text']} (Distance: {dist:.4f})")

Indices and tables