README

A Laravel package for text chunking, Ollama-based embedding generation, optional persistence, and PostgreSQL pgvector similarity search.

Current runtime support is intentionally narrow:

• Ollama embedding generation is supported.
• PostgreSQL with pgvector is required for similarity search.
• SQLite/MySQL can persist vectors, but they do not provide vector search through this package.
• OpenAI configuration is reserved for a future release and is not supported at runtime yet.

Key Features

1. Smart Context Chunking: Includes DefaultChunker, MarkdownChunker, SentenceChunker, and TokenBudgetChunker.
2. Native PostgreSQL Vector Search: Uses pgvector cosine-distance operators (<=>) when your embedding store is PostgreSQL.
3. Background Processing: Ships with queue-aware jobs plus configurable queue connection, queue name, retry/backoff, timeout, and overlap protection.
4. Safer Re-Embedding: Can skip unchanged targets and replace persisted target sets only after successful generation.
5. Flexible Targeting: Supports Eloquent-backed targets and non-Eloquent target_type / target_id references.
6. Search Helpers: Supports metadata-aware filtering and a thin EmbeddingSearch service.

Quick Start

Please read the complete documentation available in the docs/ directory:

• Installation & Setup
• Usage & Asynchronous Processing
• PostgreSQL pgvector Performance
• Dynamic Runtime Context

Basic Usage

use JOOservices\LaravelEmbedding\Facades\Embedding;
use JOOservices\LaravelEmbedding\Facades\EmbeddingSearch;

// 1. Single text raw vector
$vector = Embedding::embedText('Who is the CEO of Apple?');

// 2. Chunk, embed, and persist a non-Eloquent target
Embedding::chunkAndEmbed($hugePdfContent, [
    'target_type' => 'document',
    'target_id' => 'annual-report-2024',
    'namespace' => 'finance',
    'skip_if_unchanged' => true,
    'author' => 'System',
]);

// 3. Search & Retrieve (PostgreSQL + pgvector only)
$results = EmbeddingSearch::similarToText('Company leadership', 5, [
    'namespace' => 'finance',
    'meta' => ['author' => 'System'],
]);

PostgreSQL Notes

This package does not auto-create a pgvector ANN index because index strategy depends on your chosen model dimensions and operational preferences. Treat extension enablement and index creation as deployment decisions in the host application.

If you want the package migration to attempt CREATE EXTENSION vector, enable:

EMBEDDING_PGVECTOR_ENSURE_EXTENSION=true

AI Agents & Development

This package contains strict documentation for external AI Agents (Cursor, Cline, Github Copilot). If you are an AI Agent building on top of this package, read the Skill sheet located at .agents/skills/laravel-embedding/SKILL.md.