Laravel 13 ships stable first-party AI SDK with provider-agnostic interface

Laravel 13 has arrived with the framework's AI SDK promoted out of beta and into a stable, first-party package released alongside the framework itself. According to a post on dev.to, the package gives PHP developers a single, provider-agnostic interface for wiring language models and other AI capabilities directly into Laravel applications.

One interface, many providers

Until now, adding an LLM to a PHP project typically meant choosing among vendor-specific SDKs — OpenAI's, Anthropic's, Google's Gemini library — and coping with each one's conventions, response formats and quirks. The Laravel AI SDK collapses this into a shared abstraction: code is written once, and switching providers comes down to editing an environment variable such as AI_PROVIDER in the application's .env file. The dev.to post walks through an e-commerce scenario where SEO-friendly product descriptions are produced by configuring a provider, injecting the AI service via dependency injection, and reading back a normalized response. If the client later wants a different vendor for cost or performance reasons, the change is a single line with no refactoring and no regression risk, the post argues.

What the package covers

The SDK spans the main areas of modern AI application development, as described in the source:

• Text generation: chat and completion calls against any supported model, returning responses in a standard format regardless of the underlying provider.
• Tool-calling agents: agents that can invoke PHP functions defined in the project, autonomously selecting which tools to use toward a stated goal — suited to multi-step automation such as data analysis or report generation.
• Embeddings: numeric vectors for semantic search, clustering and retrieval-augmented generation.
• Image generation: a unified interface for creating and saving images across providers.
• Audio synthesis: integrated text-to-speech for applications needing spoken output.
• Vector stores: native integration with vector databases including Pinecone, Weaviate and pgvector for storing and retrieving embeddings efficiently.

Architecture that Laravel developers already know

Rather than introducing a new design, the package follows the framework's established patterns. A service provider registers services and configuration in the container, facades such as AI::generate() and AI::embed() give quick access, and every provider is an interchangeable driver in the same style as Laravel's cache, session and queue drivers. Advanced parameters — timeouts, retries and fallback providers — live in a config/ai.php file. Developers already fluent in Laravel therefore face no new architecture to learn, the post notes.

Testing as a first-class concern

The source highlights testability as one of the most significant benefits of first-party status. AI::fake() simulates responses without real calls to a provider, assertions can verify that specific requests were made, and predefined response sequences support testing multi-step flows. The tooling integrates with PHPUnit and Pest without extra configuration. That addresses a concrete pain point: with third-party SDKs, testing AI code often required custom wrappers, hand-built mocks or — expensively and slowly — live API calls.

Providers and extensibility

At launch the SDK is said to support OpenAI (GPT-4o, GPT-4.5, DALL-E, Whisper), Anthropic (Claude Opus, Sonnet and Haiku), Google (Gemini Pro and Ultra), Mistral (Mistral Large, Mixtral) and Ollama for self-hosted local models. Adding a custom provider means implementing an interface with a small number of methods, mirroring the driver pattern used elsewhere in the framework — relevant for teams running proprietary or niche models.

The source also counsels restraint: the SDK makes sense for text generation, semantic analysis and natural-language automation, but deterministic algorithms remain the better tool for filters, sorting and financial calculations.

Why it matters

Framework-level AI support moves LLM integration from a bolt-on dependency to a standard platform service. For the large PHP and Laravel ecosystem, it lowers the barrier to building chatbots, semantic search and agentic automation without leaving the stack or picking up Python or JavaScript tooling. The driver-based design also softens vendor lock-in, since swapping models becomes a configuration change rather than a rewrite. As with any single-source report, specifics such as the exact provider and model lineup reflect what the dev.to post describes at release time, but the broader signal is clear: AI capability is becoming table stakes for mainstream web frameworks, and Laravel is now baking it in natively.