Geocodio

Overview: The Unified AI Interface Integrating Artificial Intelligence into a modern application often feels like chasing a moving target. One week OpenAI is the leader; the next, Anthropic releases a model that changes the game. Without a proper abstraction layer, developers find themselves trapped in a maintenance nightmare, juggling multiple APIs and varying response formats. Prism solves this by providing a unified, Laravel-native routing layer for AI models. It allows you to switch providers by changing a single line of code while maintaining the fluid, expressive syntax that PHP developers expect from tools like Eloquent. Prerequisites To follow this guide, you should be comfortable with the following: * **PHP 8.2+**: Leveraging modern type-hinting and features. * **Laravel Framework**: Familiarity with Service Providers and Blade. * **AI Fundamentals**: Basic understanding of prompts, models, and API keys. Key Libraries & Tools * **Prism**: The core package providing the unified API for AI providers. * **Blade**: Used as a sophisticated templating layer for dynamic prompt engineering. * **Laravel Zero**: Mentioned for building CLI-based AI agents. * **MCP (Model Context Protocol)**: Supported via the Relay package for tool integration. Implementation: Text and Structured Output Standard text generation is the baseline. Prism uses a fluent interface to define the provider and the prompt. A standout feature is the use of **Blade templates** for prompts, which avoids the mess of concatenated strings. ```php $response = Prism::text() ->using("anthropic", "claude-3-5-sonnet") ->withPrompt(view('prompts.blog-post', ['topic' => $topic])) ->generate(); echo $response->text; ``` While text is useful, **Structured Output** is where the real power lies. Instead of using messy regex to parse markdown, you define a schema. This ensures the AI returns a predictable associative array, perfect for updating database columns directly. ```php $response = Prism::structured() ->using("openai", "gpt-4o") ->withSchema(Schema::object([ 'summary' => Schema::string(), 'links' => Schema::array(Schema::string()), ])) ->generate(); ``` Building Intelligent Agents with Tools TJ%20Miller emphasizes that AI is most effective when it can act. Prism supports **Tool Use** (function calling) where the model can decide to invoke PHP methods. By defining a class-based tool, you give the model the ability to edit files, search databases, or interact with external APIs via the Model%20Context%20Protocol. Syntax Notes and Best Practices Prism adopts the **Invocable Action** pattern, keeping controllers lean. It also supports **Multimodal** inputs, allowing you to pass images or PDFs from any Laravel-defined disk (S3, local, etc.) simply by attaching them to the request chain. This integration with the Laravel storage abstraction makes document analysis trivial. Practical Examples * **Legal Tech**: Automating contract analysis and passing summaries to human lawyers. * **Content Creation**: Enhancing sparse image descriptions into rich narratives using persona-based prompts (like the "Vivian" persona). * **DevOps Agents**: Using Laravel%20Zero to build CLI tools that can fix bugs or refactor code automatically. Tips & Gotchas * **Human in the Loop**: Always verify AI output for critical tasks like legal or financial decisions. * **Testing**: Use `Prism::fake()` to mock responses. This prevents hitting expensive APIs during your CI/CD pipeline and allows you to assert that specific prompts were sent.