Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.instructorphp.com/llms.txt

Use this file to discover all available pages before exploring further.

This guide walks you through installing Polyglot and making your first LLM inference request. By the end you will have a working PHP script that sends a prompt to an LLM provider and prints the response.
Polyglot is already included in the Instructor for PHP package. If you have Instructor installed, you do not need to install Polyglot separately.

​
Installation

Polyglot requires PHP 8.3 or later. Install it with Composer: 
composer require cognesy/instructor-polyglot
# @doctest id="0626"

​
Configure Your API Key

Polyglot ships with ready-made presets for over 25 providers including OpenAI, Anthropic, Gemini, Mistral, Groq, Deepseek, and many more. Each preset reads its API key from an environment variable so credentials never appear in code. For this quickstart we will use the openai preset. Export your key before running any PHP code: 
export OPENAI_API_KEY=sk-...
# @doctest id="c3c2"
Never hard-code API keys in source files. Use environment variables or a .env file to keep credentials out of version control.

​
Your First Request

Create a file called test-polyglot.php in your project directory: 
<?php
require __DIR__ . '/vendor/autoload.php';

use Cognesy\Polyglot\Inference\Inference;
use Cognesy\Messages\Messages;

$answer = Inference::using('openai')
    ->withMessages(Messages::fromString('What is the capital of France?'))
    ->get();

echo "ASSISTANT: $answer\n";
// @doctest id="de85"
Run it from the terminal: 
php test-polyglot.php

# Output:
# ASSISTANT: The capital of France is Paris.
# @doctest id="a134"
That is all it takes. The Inference class is the main entry point for every LLM request in Polyglot.

​
Understanding the Flow

Every Polyglot request follows a three-step pattern:
  1. Select a provider — call Inference::using('preset') to choose a bundled or custom preset, or create an Inference instance directly.
  2. Build the request — chain fluent methods such as withMessages(), withModel(), withMaxTokens(), or withOptions() to describe what you want.
  3. Execute — call a terminal method to send the request and retrieve the result.
The available terminal methods are:
MethodReturnsDescription
get()stringThe plain-text content of the response.
response()InferenceResponseThe full response object with content, usage, tool calls, and metadata.
asJson()stringThe response content parsed as a JSON string.
asJsonData()arrayThe response content parsed into a PHP array.
stream()InferenceStreamA streamed response you can iterate over in real time.

​
Switching Providers

Because every provider is just a preset name, switching from OpenAI to another provider is a one-line change: 
use Cognesy\Messages\Messages;

// Anthropic
$text = Inference::using('anthropic')
    ->withMessages(Messages::fromString('Explain dependency injection in one paragraph.'))
    ->get();

// Google Gemini
$text = Inference::using('gemini')
    ->withMessages(Messages::fromString('Explain dependency injection in one paragraph.'))
    ->get();

// Groq
$text = Inference::using('groq')
    ->withMessages(Messages::fromString('Explain dependency injection in one paragraph.'))
    ->get();
// @doctest id="3f50"
Set the corresponding environment variable for each provider you want to use (ANTHROPIC_API_KEY, GEMINI_API_KEY, GROQ_API_KEY, and so on).

​
Overriding the Model

The preset defines a default model, but you can override it per-request: 
use Cognesy\Messages\Messages;

$text = Inference::using('openai')
    ->withModel('gpt-4.1')
    ->withMessages(Messages::fromString('Summarize the theory of relativity in two sentences.'))
    ->get();
// @doctest id="dd8a"

​
Streaming Responses

For long responses or interactive UIs, you can stream the output token by token: 
use Cognesy\Messages\Messages;

$stream = Inference::using('openai')
    ->withMessages(Messages::fromString('Write a short poem about PHP.'))
    ->stream();

foreach ($stream->deltas() as $delta) {
    echo $delta->contentDelta;
}
// @doctest id="2563"

​
Next Steps

Now that you have a working setup, explore the rest of the documentation to unlock the full power of Polyglot:
  • Setup — define your own presets and customize provider configuration.
  • Essentials — learn the complete request and response API.
  • Streaming — handle streamed responses, events, and partial updates.
  • Embeddings — generate vector embeddings for semantic search and RAG.