Weave Code
Laravel Semantic Search Laravel Package
venmail/laravel-semantic-search
Natural language semantic search for Laravel Eloquent models: query in plain English, no embeddings or training. Auto-analyzes your codebase, navigates relationships, supports aggregations, multilingual input, and multi-database setups.
Laravel semantic search package with static code analysis and context-aware query building.
Frequently asked questions about Laravel Semantic Search
- What is venmail/laravel-semantic-search and how does it work?
- It is a Laravel package that lets you query Eloquent models with plain-English input. It uses static code analysis and a heuristic English-to-SQL parser instead of embeddings or model training, and it includes an optional OpenAI fallback for complex queries. It advertises relationship navigation, aggregations, multilingual input, and MySQL, PostgreSQL, or SQLite support.
- How do I install venmail/laravel-semantic-search in a Laravel app?
- Install it with composer require venmail/laravel-semantic-search, then publish config/semantic-search.php using php artisan vendor:publish --tag="semantic-search-config". The package provides a service provider and a SemanticSearch facade, and the README says model, controller, view, and route discovery is automatic. Pin an exact version in composer.json because the package is at 0.x and has already shipped breaking changes.
- Which Laravel and PHP versions does laravel-semantic-search support?
- The README lists PHP 8.1 or higher and Laravel 9.x, 10.x, or 11.x, with MySQL 8.0+, PostgreSQL 12+, or SQLite 3.0+. Laravel 12 support is not stated, and with zero dependents there is little ecosystem pressure to add it quickly. Verify your exact database and Laravel minor version in a sandbox before adopting.
- Does laravel-semantic-search require embeddings, a vector database, or model training?
- No. The package's core value proposition is static code analysis plus a heuristic parser, so there is no embedding pipeline, vector database, GPU spend, or training step. Enabling the optional LLM fallback does reintroduce an external OpenAI dependency with its own latency, cost, and compliance profile.
- Can it handle complex queries with Eloquent relationships and aggregations?
- The package advertises relationship intelligence, aggregation support, and boolean logic, so it can attempt those queries. However, multi-clause natural-language queries are exactly where heuristic parsers are least reliable, and the README's headline examples may not work consistently without the LLM fallback. Test a corpus of your real queries rather than assuming complex reporting queries will parse correctly.
- Is venmail/laravel-semantic-search production-ready?
- Treat the README's production-ready claim with caution. Observable signals include version 0.3.0, one star, zero dependents, a breaking class rename, no visible CI or coverage badges, and no security policy. For most teams this is a greenfield spike candidate, not a first production search layer, and you should assume you would be an early production user.
- Does the generated Eloquent query respect global scopes, policies, and multi-tenancy?
- This is unclear from the available material and should be resolved before any commitment. The security config uses blocklists such as restricted_models and restricted_fields, which fail open for any new model or column unless an allowlist mode exists. Ask specifically whether generated Eloquent queries respect global scopes, wherePolicies, and tenant scoping; if the answer is unclear, treat it as a blocker for multi-tenant apps.
- What happens when the parser cannot understand or ambiguously parses a natural language query?
- The key risk is whether the parser fails loudly or silently. Silent misparsing that returns plausible but wrong rows is the most dangerous failure mode for reporting, billing, or compliance surfaces. Test ambiguous and unsupported queries explicitly, and do not rely on the package for critical results unless you can detect and reject bad parses.
- What data is sent to OpenAI when the LLM fallback is enabled?
- The package documentation does not specify exactly what is sent to OpenAI when LLM fallback is enabled. It could include query text, schema details, or column names, which makes this a compliance and data-handling question rather than just a code question. Keep the fallback disabled until your security or legal team reviews the payload.
- How should I test and pin versions before adopting this package in production?
- Spike it in an isolated branch against a read replica or sandbox database, never production data. Build a corpus of 50 to 100 real natural-language queries with expected results, then test relationships, aggregations, ambiguous queries, tenant scoping, cache invalidation, and read-only container filesystems. Pin an exact version in composer.json and treat each 0.x upgrade as a mini-migration because public classes have already changed.
Weaver
How can I help you explore Laravel packages today?