database

Database & Embeddings

Supabase Client

Use the configured client from src/db/client.js:

import { supabase, isSupabaseConfigured } from '../db/client.js';

if (!isSupabaseConfigured()) {
  // Handle gracefully
}

Embedding Providers

Two providers supported (NOT interchangeable):

Provider	Model	Dimensions
OpenAI	text-embedding-3-small	1536
Ollama	nomic-embed-text	1024

Check configuration before use:

import { isOpenAIConfigured, generateEmbedding } from '../services/embeddings.js';

if (!isOpenAIConfigured()) {
  // Return error to user
}

const embedding = await generateEmbedding(text);

Text Chunking

Use the chunker service for consistent chunking:

import { chunkText } from '../services/chunker.js';

const chunks = chunkText(content);
// Returns: { content, index, startOffset, endOffset, tokenCount }[]

Chunking parameters:

• Max chunk size: 512 tokens (~2048 chars)
• Overlap: 50 tokens
• Strategy: Paragraph-aware (splits on \n\n)

Hybrid Search

Use hybridSearch() for combined FTS + semantic search:

import { hybridSearch } from '../db/search.js';

const results = await hybridSearch({
  queryText: query,
  queryEmbedding: embedding,
  limit: 10,
  fullTextWeight: 1.0,
  semanticWeight: 1.0,
});

Returns results ranked by Reciprocal Rank Fusion (RRF).

Database Schema

Key tables:

• documents: Full documents with tsvector for FTS
• chunks: Document chunks with vector[1536] embeddings (HNSW index)

RPC functions:

• hybrid_search(): Performs RRF fusion of FTS + vector results