Deploy and Host vchord-suite on Railway

VectorChord Suite is a PostgreSQL distribution with built-in vector search and BM25 full-text search. It includes vchord for IVF+RaBitQ vector indexes that scale to billions of vectors, pg_tokenizer for text tokenization, and vchord_bm25 for hybrid search—all with pgvector compatibility.

About Hosting vchord-suite

Hosting VectorChord Suite requires a PostgreSQL-compatible environment with persistent storage for your vector data. This template pre-configures all four extensions (vchord, pg_tokenizer, vchord_bm25, vector) and initializes them on first boot. Unlike standard pgvector HNSW indexes which require significant RAM, VectorChord's IVF+RaBitQ indexes use disk-based quantization, making it cost-effective to scale to millions of vectors without expensive memory upgrades.

Common Use Cases

• AI/ML applications — Semantic search, RAG (retrieval-augmented generation), recommendation systems
• Hybrid search — Combine vector similarity with BM25 keyword search for better relevance
• Large-scale embeddings — Store and query 1M+ vectors without hitting memory limits

Dependencies for vchord-suite Hosting

• PostgreSQL 17
• Persistent volume for database storage

Deployment Dependencies

• VectorChord Documentation
• TensorChord Docker Hub
• VectorChord GitHub

Implementation Details

-- Create a scalable vector index
CREATE INDEX ON items USING vchordrq (embedding vector_ip_ops)
WITH (options = $$
residual_quantization = true
[build.internal]
lists = [1000]
$$);

-- Configure recall vs latency (set probes to 3-10% of lists)
ALTER DATABASE railway SET vchordrq.probes = 50;