README

Official Ollama provider for the framework-agnostic PHP AI SDK. It supports text, streaming, native embeddings, and Ollama's experimental OpenAI-compatible image endpoint.

Installation

composer require aisdk/ollama

Basic Usage

use AiSdk\Generate;
use AiSdk\Ollama;

$result = Generate::text()
    ->model(Ollama::model('your-installed-model'))
    ->prompt('Explain closures in PHP.')
    ->run();

echo $result->text;

Any model installed on the configured Ollama server can be selected. Like every AI SDK provider package, this package treats model IDs as opaque provider values and does not ship a model inventory.

Ollama's OpenAI-compatible Chat Completions endpoint is the default. Select the Responses endpoint globally or per request when needed:

Ollama::create(['api' => 'responses']);

$result = Generate::text('Explain this code.')
    ->model(Ollama::model('your-installed-model'))
    ->providerOptions('ollama', ['api' => 'responses'])
    ->run();

Supported values are chat_completions and responses. Chat Completions supports Ollama's documented vision, function tools, reasoning effort, and structured-output fields. Ollama Responses supports streaming and function tools, but its documented request fields do not currently expose reasoning or structured-output controls. The selected installed model remains responsible for accepting model-dependent features.

Available models

Read the current server registry instead of relying on package data:

$models = Ollama::availableModels();

foreach ($models as $model) {
    echo $model->id;
}

availableModels() calls Ollama's native /api/tags endpoint. To load a model's optional capabilities—such as embeddings, vision, tools, audio input, or thinking—inspect it explicitly:

$definition = Ollama::inspectModel('your-installed-model');

if (in_array('image_input', $definition->capabilityNames(), true)) {
    // The installed model advertises vision support.
}

Inspection calls /api/show and returns informational details reported by the current Ollama server. It does not change adapter behavior, and ordinary generation does not perform hidden discovery requests.

Embeddings

Ollama embeddings use the native /api/embed endpoint and accept one text or a batch:

use AiSdk\Generate;
use AiSdk\Ollama;

$result = Generate::embedding(['Search query', 'Document text'])
    ->model(Ollama::embedding('embeddinggemma'))
    ->dimensions(256)
    ->providerOptions('ollama', [
        'truncate' => false,
        'keep_alive' => '10m',
    ])
    ->run();

$queryVector = $result->embeddings[0]->vector;
$documentVector = $result->embeddings[1]->vector;

truncate and keep_alive are passed to Ollama unchanged. The selected installed model remains the authority on supported output dimensions.

Image generation

$image = Generate::image('A small robot')
    ->model(Ollama::image('your-installed-image-model'))
    ->size('1024x1024')
    ->run();

Image generation is experimental in Ollama. Ollama does not currently expose an OpenAI-compatible speech endpoint, so this package intentionally does not implement Generate::speech().

Streaming

use AiSdk\Generate;
use AiSdk\Ollama;

foreach (Generate::text('Tell me a story.')->model(Ollama::model('your-installed-model'))->stream()->chunks() as $chunk) {
    echo $chunk;
}

Configuration

No API key is required for a local Ollama server. Provide one only if your server is protected.

Variable	Description	Default
`OLLAMA_BASE_URL`	Base URL for the OpenAI-compatible API	`http://localhost:11434/v1`
`OLLAMA_NATIVE_BASE_URL`	Base URL for native API endpoints	Derived by removing `/v1` from `OLLAMA_BASE_URL`
`OLLAMA_API_KEY`	Optional bearer token	—

Ollama::create([
    'baseUrl' => 'http://localhost:11434/v1',
    'nativeBaseUrl' => 'http://localhost:11434',
]);

Testing

composer test

aisdk / ollama

Maintainers

Package info

Statistics

Security