Taylor Otwell has been working on Laravel AI SDK for a few months now, and at Laracon India 2026, he showed the demo.
The actual public release should be next week, but I took many screenshots, and now, back at the hotel, I wanna show you what the features will be.
Wait, What is AI SDK?
In general, the goal is to call AI providers for various operations:
- Get general "chatbot" answers like ChatGPT
- Get search answers based on your database via embeddings
- Generate video/audio and transcriptions
- ... and many more
And Taylor's idea is to provide an elegant Laravel syntax and API, to make it very easy to use, no matter which AI provider you choose.
Basically, you just call something like agent()->prompt('your query here...') and get responses.
But it goes much deeper. Let's see the examples.
First, Configure AI Providers
Here are a few screenshots, how you will configure your API keys from Anthropic, Open AI and others.
config/ai.php:
And then you use provider name like 'anthropic' in the Agent prompts, as you'll see below.
Convenient, isn't it?
Basic Usage: Prompt Agent for Answers
As mentioned above, you may just prompt the agent with agent() helper, like this:
And this is a typical response structure:
JsonSchema for Custom Structure
You may provide the schema how you want the results returned:
And here's the response example Taylor showed on stage:
Queueing and Streaming AI Responses
We all know that LLM answers may take time. So, you can offload the agent prompt to a queue, or choose to stream the response, like in a typical chatbot, word by word, as it gets returned.
Side note: that notify() function is a custom helper Taylor wrote for the demo purposes, it's not included in AI SDK.
Generating Image (Also With Queues)
Here's an example where batteries-included Laravel approach shines: you can use new AI SDK features, combined with Laravel core flexible features like queues and filesystem.
Here's how you generate an image with, for example, Nanobanana API, and store it locally.
Also, you can generate a new image based on existing image, adding a prompt to AI.
Audio and Transcriptions
Similar to the images above, you can work with audio, using provider like ElevenLabs:
Embeddings and Similarity Search
With new AI SDK, it will be easy to introduce search like this into your project:
Notice the search query is about "big boats", but the movie "Titanic" description doesn't contain the word "boat". That's search with the help of embeddings.
It's actually pretty complex under the hood, but here's the surface-level code in a Controller, for example:
With this feature released, I will probably re-shoot this recent video: Laravel AI Chatbot Trained on Your Data: Example Project, but you can still watch it, if you want to understand Embeddings.
Also, I'm not 100% sure, but I think this feature will work best in PostgreSQL, because it has native vector search functionality, recently added to Laravel 12.
Agent Classes
You will also be able to run php artisan make:agent and have a dedicated class:
Agent classes may have many options to configure, like UseCheapestModel attribute, which will default the model to -haiku, -nano, and similar models from AI providers.
Taylor also briefly showed many more things you can add to your agents, like:
- Middleware
- Personalization
- Structure
- Tools with schemas
- Web search
- etc.
I will talk about them more in the future review, when SDK is officially out.
For now, here's a link to a Google Drive folder, if you want to see more screenshots (sorry for the quality).
I'm excited for the AI SDK, and will probably shoot a short course about it, with realistic project examples, later in February. Wanna know when it's out? Subscribe to my weekly newsletter :)