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.

OllamaEmbeddingModel implements EmbeddingModelInterface and sends requests to Ollama’s native /api/embed endpoint — not the OpenAI-compatible path. It supports both single and batch inputs, optional dimension control, and provider-specific pass-through options such as truncate and keep_alive.

​
Namespace

use AiSdk\Ollama\Models\OllamaEmbeddingModel;
This class is not instantiated directly. Use Ollama::model() to obtain a model instance and the Generate facade to dispatch requests.

​
URL

Requests are sent to: 
{nativeBaseUrl}/api/embed
The nativeBaseUrl defaults to http://localhost:11434 and is distinct from the OpenAI-compatible baseUrl (http://localhost:11434/v1) used by text and image models. It can be configured independently: 
Ollama::create([
    'baseUrl'       => 'http://my-ollama-host:11434/v1',
    'nativeBaseUrl' => 'http://my-ollama-host:11434',
]);
If nativeBaseUrl is not set explicitly, it is derived automatically by stripping the /v1 suffix from baseUrl.

​
Method

​
generate(EmbeddingRequest $request): EmbeddingResponse

Builds a request body and POSTs it to /api/embed, then validates and parses the response into an EmbeddingResponse.

​
Request body

The following fields are sent to Ollama:
​
model
string
required
The model identifier string passed at construction — for example 'nomic-embed-text'.
​
input
string[]
required
Array of input strings to embed, sourced from $request->inputs. Ollama processes all inputs in a single batch call.
​
dimensions
int
Optional number of dimensions to truncate the output vectors to. Sourced from $request->dimensions. Omitted from the request body if not set.

​
Provider options merging

Any provider options passed via ->providerOptions('ollama', [...]) are merged into the request body using array_replace_recursive, allowing you to set Ollama-specific fields:
OptionTypeDescription
truncateboolWhether to truncate inputs to the model’s context window. Defaults to true in Ollama.
keep_alivestringDuration to keep the model loaded in memory, e.g. '10m' or '1h'.
The special raw key inside provider options is also merged as a final passthrough, applied after all other fields. Use it to forward any undocumented or future Ollama fields without interference from the SDK’s merge logic.

​
Response Fields

​
embeddings
EmbeddingData[]
One EmbeddingData object per input string, in the same order as the input array.
​
embeddings[n].vector
float[]
The embedding vector for the nth input. The number of dimensions matches the model’s native output size, or the requested dimensions value if truncation was applied.
​
usage.inputTokens
int
Total number of tokens processed across all inputs, sourced from prompt_eval_count in the Ollama response. Returns 0 if Ollama does not include token counts.
​
providerMetadata.ollama.model
string
The model string echoed back by Ollama in the response. Falls back to the requested model ID if the field is absent.
​
providerMetadata.ollama.total_duration
int
Total time taken for the request in nanoseconds, as reported by Ollama.
​
providerMetadata.ollama.load_duration
int
Time spent loading the model into memory in nanoseconds. Only present when Ollama includes this field.

​
Error Conditions

generate() raises an InvalidResponseException in two cases: Unexpected embedding count — if the number of embedding vectors in the Ollama response does not match the number of input strings: 
Ollama returned an unexpected number of embeddings.
This can occur if Ollama silently drops inputs or returns a malformed response. Invalid embedding vector — if any individual embedding vector is empty or contains non-numeric values: 
Ollama returned an invalid embedding.
The exception context includes the inputIndex of the offending vector to aid debugging.

​
Usage Example

use AiSdk\Generate;
use AiSdk\Ollama;

Ollama::create();

$result = Generate::embedding(['First document', 'Second document'])
    ->model(Ollama::model('nomic-embed-text'))
    ->dimensions(256)
    ->providerOptions('ollama', [
        'truncate'   => false,
        'keep_alive' => '10m',
    ])
    ->run();

foreach ($result->embeddings as $embedding) {
    echo 'Vector dimensions: ' . count($embedding->vector) . PHP_EOL;
}

​
Accessing provider metadata

Ollama::create();

$result = Generate::embedding(['Hello world'])
    ->model(Ollama::model('nomic-embed-text'))
    ->run();

$meta = $result->providerMetadata['ollama'];

echo 'Model: '          . $meta['model']          . PHP_EOL;
echo 'Total duration: ' . $meta['total_duration'] . ' ns' . PHP_EOL;
echo 'Load duration: '  . $meta['load_duration']  . ' ns' . PHP_EOL;
echo 'Input tokens: '   . $result->usage->inputTokens . PHP_EOL;

​
Using the raw passthrough

$result = Generate::embedding(['Document text here'])
    ->model(Ollama::model('nomic-embed-text'))
    ->providerOptions('ollama', [
        'keep_alive' => '5m',
        'raw' => [
            'some_future_ollama_field' => 'value',
        ],
    ])
    ->run();

Build docs developers (and LLMs) love

Get started for freeTalk to us