Ollama PHP

A lightweight PHP client for the Ollama API.

🦙 + 🐘 = ❤️

---

Installation

composer require ollama/php --prefer-dist

Quick Start

Get up and running with the Ollama PHP client in two lines of code. The client ships with two entry points: a fluent Ollama facade that returns typed response objects, and a low-level Client that gives you direct access to the HTTP API.

Using the Ollama facade

The Ollama class is the recommended starting point. It exposes every API endpoint as a chainable method (chat(), completions(), embed(), models(), blobs(), systemOne()) and hydrates every response into a dedicated DTO, so you get IDE autocompletion, proper types, and no array-key guessing.

use Ollama\Ollama;

$ollama = new Ollama;

$response = $ollama->chat()->send([
    'model' => 'gemma4:12b',
    'messages' => [
        ['role' => 'user', 'content' => 'Hello world!'],
    ],
]);

echo $response->message->content;

Here $response is a ChatResponse instance - $response->message->content is a typed string, not $response['message']['content']. Use this when you want the ergonomics of an SDK.

Using the Client directly

If you prefer to talk to Ollama's HTTP API yourself - for example, to hit an endpoint not yet wrapped by the SDK, or to inspect the raw JSON - instantiate Client and call post() / get() / delete() yourself. Requests and responses are plain arrays.

use Ollama\Client;

$client = new Client;

$result = $client->post('api/chat', [
    'model' => 'gemma4:12b',
    'messages' => [
        ['role' => 'user', 'content' => 'Hello world!'],
    ],
]);

var_dump($result);

This is the same request, but you get the raw associative array back - no DTO, no magic. Use this when you want full control over the payload or need to reach an endpoint the SDK doesn't yet cover.

Configuring the client

By default both entry points talk to http://127.0.0.1:11434. To point at a different host (remote server, Ollama Cloud, a proxy), pass options to the constructor:

$ollama = new Ollama(
    host: 'https://ollama.com',
    apiKey: getenv('OLLAMA_API_KEY'),
    httpOptions: [] // GuzzleHttp\Client
);

Streaming

Both Completions and Chat expose a stream() method that returns a StreamResponse you can iterate over to read chunks as they arrive:

foreach ($ollama->chat()->stream([/* ... */]) as $chunk) {
    echo $chunk->message->content;
}

Examples 👀

Every endpoint has a runnable example in the examples directory - from basic chat and completions to streaming, embeddings, and model management. Browse the folder to see each one in context.

License

MIT