# Aurabase Vector Search & RAG Reference (pgvector)

> Documentation: https://aurabase.cloud/en/docs/guides/rag-pgvector
> Engine: PostgreSQL 18 with `vector` (`pgvector`) extension

## Technical Specifications
- **Extension**: `vector` enabled on all Aurabase clusters.
- **Standard Dimensions Supported**:
  - 1536: OpenAI `text-embedding-3-small`, `text-embedding-ada-002`
  - 3072: OpenAI `text-embedding-3-large`
  - 768: Google Gemini `text-embedding-004`, Mistral `mistral-embed`
- **Distance Operators**:
  - `<=>`: Cosine Distance (recommended for semantic search)
  - `<->`: Euclidean Distance (L2)
  - `<#>`: Negative Inner Product

## Table & High-Performance HNSW Index
```sql
create extension if not exists vector;

create table documents (
  id        uuid primary key default gen_random_uuid(),
  content   text not null,
  metadata  jsonb default '{}'::jsonb,
  embedding vector(1536) not null
);

-- HNSW index for sub-5ms p99 similarity search
create index idx_documents_embedding on documents
using hnsw (embedding vector_cosine_ops)
with (m = 16, ef_construction = 64);
```

## Cosine Similarity RPC Search Function
```sql
create or replace function match_documents (
  query_embedding vector(1536),
  match_threshold float,
  match_count int
)
returns table (id uuid, content text, metadata jsonb, similarity float)
language sql stable security invoker as $$
  select id, content, metadata, 1 - (embedding <=> query_embedding) as similarity
  from documents
  where 1 - (embedding <=> query_embedding) > match_threshold
  order by embedding <=> query_embedding
  limit match_count;
$$;
```

## AI Gateway Integration (Sub-10ms Vector Pipelines)
Instead of manually generating embeddings and querying raw SQL, the Aurabase AI Gateway automates the entire vector lifecycle:
```typescript
// 1. Automatic Ingestion & Chunking into pgvector
await aura.ai.ragIngest({
  document: 'Enterprise knowledge base document text...',
  namespace: 'kb',
  chunk_size: 1000,
  chunk_overlap: 200,
})

// 2. High-Performance Semantic Search
const { data: results } = await aura.ai.semanticSearch(
  'What is our refund policy?',
  'kb',
  { threshold: 0.82, top_k: 5 }
)

// 3. One-Call Grounded RAG Synthesis
const { data: rag } = await aura.ai.rag({
  question: 'What is our refund policy?',
  namespace: 'kb',
})
console.log(rag?.answer, rag?.sources)
```