Getting Started with NeuronDB
Introduction
📌 Branch & Version Selection
NeuronDB has three branches with different versions. Choose based on your needs:
| Branch | Version | Status | Use When |
|---|---|---|---|
main | 3.0.0-devel | Latest | New projects, development, latest features (default) |
REL2_STABLE | 2.0.0 | Stable | Production, stable v2.0 features |
REL1_STABLE | 1.0.0 | Stable | Production, maximum stability required |
Note: This documentation reflects version 3.0.0-devel from the main branch. For stable releases, use REL2_STABLE (v2.0.0) or REL1_STABLE (v1.0.0). See GitHub repository for branch details.
NeuronDB is a PostgreSQL AI ecosystem that provides GPU-accelerated vector search, ML inference, hybrid retrieval, and complete agent infrastructure. The ecosystem includes:
- NeuronDB - PostgreSQL extension with vector search, ONNX model inference, and GPU acceleration
- NeuronAgent - REST API and WebSocket agent runtime with long-term memory and tool execution
- NeuronMCP - Model Context Protocol server with 100+ tools for MCP-compatible clients (like Claude Desktop)
- NeuronDesktop - Unified web interface for managing all components
What you can build: Semantic search, RAG pipelines, AI agents with PostgreSQL-backed memory, and MCP integrations - all in one unified ecosystem.
Choose Your Path
Pick the installation method that best fits your needs:
| Method | Best For | Time | Difficulty |
|---|---|---|---|
| Simple Start | Beginners, fastest setup | 5 minutes | ⭐ Easy |
| Docker Quick Start | Complete ecosystem, Docker users | 5 minutes | ⭐ Easy |
| Quick Start Guide | Technical users, first queries | 10 minutes | ⭐⭐ Medium |
| Source Build | Production, custom builds, developers | 30+ minutes | ⭐⭐⭐ Advanced |
💡 Note: New here? Use Simple Start for a beginner-friendly guide. For the complete ecosystem with Docker, use Docker Quick Start. Technical users can use Quick Start Guide.
Docker Quick Start
Complete NeuronDB ecosystem running in under 5 minutes with Docker Compose. This method includes all components with GPU support (CUDA, ROCm, Metal) and requires no manual configuration.
Start complete ecosystem
# Clone repository (main branch = 3.0.0-devel)
git clone https://github.com/neurondb-ai/neurondb.git
cd neurondb
# For stable 1.0.0 release, checkout REL1_STABLE branch:
# git checkout REL1_STABLE
# Start all services
docker compose up -d
# Verify services
docker compose psThis starts:
- NeuronDB (PostgreSQL with extension) on port 5433
- NeuronAgent (REST API) on port 8080
- NeuronMCP (MCP server)
- NeuronDesktop (Web UI) on port 3000
Source Build (Advanced)
For production deployments or custom builds, install from source. This requires PostgreSQL 16-18, a C toolchain, and build dependencies.
See the Installation Guide for detailed platform-specific instructions (Ubuntu/Debian, macOS, Rocky Linux/RHEL).
Quick reference (Ubuntu/Debian)
sudo apt-get install -y postgresql-17 postgresql-server-dev-17 build-essential
git clone https://github.com/neurondb-ai/neurondb.git
cd neurondb
# For stable 1.0.0 release, checkout REL1_STABLE branch:
# git checkout REL1_STABLE
cd NeuronDB
make PG_CONFIG=/usr/lib/postgresql/17/bin/pg_config
sudo make install PG_CONFIG=/usr/lib/postgresql/17/bin/pg_configAfter installing the extension, you'll need to separately build and run NeuronAgent, NeuronMCP, and NeuronDesktop. See component documentation for details.
Next Steps
After installation, use these guides:
- SQL Recipe Library - Ready-to-run SQL queries for vector search, hybrid search, and more
- Quick Start Guide - Create your first vector table, generate embeddings, and run semantic search
- NeuronAgent - Build AI agents with REST API and WebSocket
- NeuronMCP - Use 100+ MCP tools with Claude Desktop
- Vector Indexing - Configure HNSW, IVF, and quantization
- RAG Pipelines - Build retrieval augmented generation workflows