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.

Streaming lets you process model output incrementally as it is generated, rather than waiting for the full response. The Ollama provider delivers chunks as server-sent events over a streaming HTTP connection. Both the Chat Completions and Responses APIs support streaming, so you can use it regardless of which endpoint is configured.

​
Basic streaming

Call stream() instead of run(), then iterate over chunks(): 
use AiSdk\Generate;
use AiSdk\Ollama;

foreach (Generate::text('Tell me a story.')
    ->model(Ollama::model('llama3.2'))
    ->stream()
    ->chunks() as $chunk) {
    echo $chunk;
}
Each $chunk is a plain string containing the next piece of text yielded by the model.

​
Streaming with the Responses API

Streaming works the same way when using the Responses endpoint. Switch globally and use stream() as usual: 
Ollama::create(['api' => 'responses']);

foreach (Generate::text('Summarize this document.')
    ->model(Ollama::model('llama3.2'))
    ->stream()
    ->chunks() as $chunk) {
    echo $chunk;
}

​
How the two parsers differ

Under the hood, the adapter selects a different stream parser depending on the configured API:
  • Chat Completions — uses ChatStreamParser, which parses the data: lines from Ollama’s /chat/completions stream and yields each text delta as a string.
  • Responses API — uses ResponsesStreamParser, which parses the event stream from Ollama’s /responses endpoint and yields each text delta as a string.
Both parsers expose the same PHP Generator interface through chunks(), so your application code is identical regardless of which parser is active.
Streaming is one of the declared ADAPTER_CAPABILITIES in OllamaTextModel, which means it is always available from the adapter’s perspective. Whether the installed model streams correctly depends on the model itself.
When streaming inside a web request (for example, in a long-polling or SSE handler), PHP may buffer output before sending it to the browser. Call ob_flush() and flush() inside the loop body to push each chunk to the client immediately:
foreach (Generate::text('Tell me a story.')
    ->model(Ollama::model('llama3.2'))
    ->stream()
    ->chunks() as $chunk) {
    echo $chunk;
    ob_flush();
    flush();
}

Build docs developers (and LLMs) love

Get started for freeTalk to us