NeuronDB is a PostgreSQL extension that brings AI workloads into the database. It supports vector search, machine learning, and agent execution while keeping data inside PostgreSQL. The system is built as an ecosystem of four components that run independently but share the same database instance.

Core Components

• NeuronDB
  A PostgreSQL extension that provides vector search, hybrid search, and machine learning functions as native SQL. All operations run inside the database.

• NeuronAgent
  A runtime service that exposes REST and WebSocket APIs for AI agents. It supports task execution, memory storage, and coordination using PostgreSQL as the backend.

• NeuronMCP
  A Model Context Protocol server that allows MCP-compatible clients to access NeuronDB features through standard protocols.

• NeuronDesktop
  A web interface for managing data, models, agents, and system state across the NeuronDB ecosystem.

Features

• Native PostgreSQL Integration
  Use existing SQL skills and infrastructure. No external data movement.

• Machine Learning in SQL
  More than 50 ML functions for classification, regression, clustering, and analytics.

• Hybrid Querying
  Combine vector similarity search with full-text search for precise retrieval.

• GPU Support
  Accelerate indexing and computation with CUDA, ROCm, and Metal.

• Flexible Deployment
  Each component runs independently or together, based on system needs.

Data Flow

Clients connect through NeuronDesktop, NeuronAgent, or NeuronMCP. Requests execute SQL, vector, or ML operations inside PostgreSQL via NeuronDB. Results return through the service layer to clients.

Use Cases

• Semantic and similarity search
• Retrieval-Augmented Generation pipelines
• Agent-based systems with memory and task execution

Getting Started

Documentation, tutorials, and APIs are available at
https://www.neurondb.ai/docs

Website

https://www.neurondb.ai