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.

The aisdk/ollama package ships no model inventory. Model IDs are opaque provider values — any string your Ollama server recognises is a valid argument to Ollama::model(). To know which models are actually installed and what they support, query the live server using availableModels() and inspectModel().

​
Listing available models

availableModels() calls Ollama’s native /api/tags endpoint and returns a ModelDefinition[] array: 
use AiSdk\Ollama;

$models = Ollama::availableModels();

foreach ($models as $model) {
    echo $model->id . PHP_EOL;
}
Each ModelDefinition in the array includes a metadata array with the following fields when present in the server response:
FieldDescription
modified_atISO 8601 timestamp of when the model was last modified.
sizeModel file size in bytes.
digestContent digest (e.g. sha256:abc123…).
detailsNested object with fields such as family, parameter_size, quantization_level, etc.

​
Inspecting a model

inspectModel() calls Ollama’s /api/show endpoint and returns a single ModelDefinition populated with the capabilities the installed model advertises: 
$definition = Ollama::inspectModel('llava');

if (in_array('image_input', $definition->capabilityNames(), true)) {
    // Model advertises vision support
}
The returned ModelDefinition also includes a metadata['ollama_capabilities'] array containing the raw Ollama capability strings as returned by the server.

​
Capability mapping

OllamaProvider::mapCapabilities() translates Ollama’s native capability strings into PHP AI SDK Capability enum values:
Ollama capabilitySDK Capability
completionTextGeneration, Streaming, TextInput
visionImageInput
audioAudioInput
toolsToolCalling
thinkingReasoning
image / image_generationImageGeneration
embeddingEmbedding

​
Adapted capabilities

When a model reports the completion capability, the provider also adds a structured_output adapted capability entry with the strategy JSON schema passed through the response format. This reflects that structured output is available on Chat Completions via the response_format field, even though Ollama does not report it as a native capability string.

​
Inspection does not affect generation

Calling inspectModel() is entirely informational. It does not change how the adapter behaves during generation calls, and ordinary Generate::text(), Generate::embedding(), and Generate::image() calls never make hidden discovery requests to /api/show or /api/tags.
Both availableModels() and inspectModel() use nativeBaseUrl (default: http://localhost:11434) to construct their request URLs. See the configuration options page to learn how to customise nativeBaseUrl for remote or non-standard Ollama deployments.

Build docs developers (and LLMs) love

Get started for freeTalk to us