Llm Providers

Instructor is provider-agnostic. Provider selection is handled through LLMConfig and the Polyglot runtime that powers the underlying inference calls. Once the configuration is resolved, your application code stays the same regardless of which provider you use.

Selecting a Provider

The most common approaches:

use Cognesy\Instructor\StructuredOutput;
use Cognesy\Polyglot\Inference\Config\LLMConfig;

// Use a named preset
$so = StructuredOutput::using('anthropic');

// Build from a config object
$so = StructuredOutput::fromConfig(
    LLMConfig::fromPreset('openai')
);

Presets are YAML files stored in the config/llm/presets directory. Each file defines the API URL, driver, default model, and other provider-specific settings.

Supported Providers

The following providers have built-in presets:

Provider	Preset name
A21	a21
Anthropic	anthropic
AWS Bedrock	aws-bedrock
Azure OpenAI	azure
Cerebras	cerebras
Cohere	cohere
DeepSeek	deepseek
DeepSeek (Reasoning)	deepseek-r
Fireworks	fireworks
Google Gemini	gemini
Gemini (OpenAI-compatible)	gemini-oai
GLM	glm
Groq	groq
Hugging Face	huggingface
Inception	inception
Meta	meta
MiniMaxi	minimaxi
MiniMaxi (OpenAI-compatible)	minimaxi-oai
Mistral	mistral
Moonshot / Kimi	moonshot-kimi
Ollama	ollama
OpenAI	openai
OpenAI Responses	openai-responses
OpenRouter	openrouter
Perplexity	perplexity
Qwen	qwen
SambaNova	sambanova
Together	together
xAI	xai

Custom Providers

Any OpenAI-compatible API can be used by building an LLMConfig manually:

use Cognesy\Polyglot\Inference\Config\LLMConfig;
use Cognesy\Instructor\StructuredOutput;

$config = new LLMConfig(
    apiUrl: 'https://my-provider.example.com/v1',
    apiKey: $_ENV['MY_PROVIDER_KEY'],
    model: 'my-model',
    driver: 'openai-compatible',
    maxTokens: 2048,
);

$result = StructuredOutput::fromConfig($config)
    ->with(
        messages: 'Extract the data.',
        responseModel: MyModel::class,
    )
    ->get();