Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/phpaisdk/ollama/llms.txt

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

This guide covers how to generate text with the aisdk/ollama provider. It works with any model installed on your Ollama server — the package treats model IDs as opaque values and ships no model inventory of its own. Use Ollama::model() to reference any identifier your server recognises.

​
Basic text generation

use AiSdk\Generate;
use AiSdk\Ollama;

$result = Generate::text()
    ->model(Ollama::model('llama3.2'))
    ->prompt('Explain closures in PHP.')
    ->run();

echo $result->text;

​
Choosing an API endpoint

Ollama exposes two OpenAI-compatible API surfaces. The default is OllamaApi::ChatCompletions (chat_completions). You can switch to the Responses API globally when creating the provider: 
Ollama::create(['api' => 'responses']);
Or override the endpoint on a per-request basis using providerOptions(): 
$result = Generate::text('Explain this code.')
    ->model(Ollama::model('llama3.2'))
    ->providerOptions('ollama', ['api' => 'responses'])
    ->run();
Valid values are chat_completions and responses. The Chat Completions endpoint supports vision, reasoning effort, and structured output. The Responses endpoint supports streaming and tool calling, but does not expose reasoning or structured-output controls.

​
Vision (image input)

Image input is supported on the Chat Completions endpoint only. Pass a Message with mixed content using Content::text() and Content::image(): 
use AiSdk\Content;
use AiSdk\Message;

$result = Generate::text()
    ->model(Ollama::model('llava'))
    ->messages([Message::user([
        Content::text('Describe this image'),
        Content::image('https://example.com/photo.png'),
    ])])
    ->run();

​
Reasoning effort

Reasoning effort is supported on the Chat Completions endpoint only. Pass a Reasoning instance to the reasoning() method: 
use AiSdk\Reasoning;

$result = Generate::text('Solve this problem step by step.')
    ->model(Ollama::model('qwq'))
    ->reasoning(Reasoning::effort('low'))
    ->run();
The minimal effort level is not supported by Ollama. Valid values are low, medium, and high. Passing minimal throws an InvalidArgumentException.
Reasoning is not supported on the Responses API. Using reasoning() together with providerOptions('ollama', ['api' => 'responses']) throws an InvalidArgumentException at runtime.

​
Structured output

Structured output is supported on the Chat Completions endpoint only. It works by passing a JSON schema through the response_format field.
Structured output is not available on the Responses API. Using a structured output configuration with the Responses endpoint throws an InvalidArgumentException at runtime.

​
Adapter capabilities

The adapter declares a fixed set of ADAPTER_CAPABILITIES in OllamaTextModel: TextGeneration, Streaming, ToolCalling, StructuredOutput, Reasoning, TextInput, and ImageInput. These reflect what the adapter code supports, not what any particular model supports. The installed model on your Ollama server is still the authority on which features actually work at inference time.

Build docs developers (and LLMs) love

Get started for freeTalk to us