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 Ollama provider exposes two API surfaces: Chat Completions (/chat/completions) and Responses (/responses). Both surfaces are OpenAI-compatible and support streaming and tool calling. Chat Completions is the default because it has broader documented support for Ollama-specific features — vision input, reasoning effort controls, and structured output. The Responses surface is available when you need it, but certain features that rely on Chat Completions request fields are intentionally blocked and will throw at request time.

​
Endpoint comparison

FeatureChat CompletionsResponses
Default✓—
Streaming✓✓
Tool calling✓✓
Vision (image input)✓— (not documented)
Reasoning effort✓ (low, medium, high)✗ (throws)
Structured output✓✗ (throws)
URL path/chat/completions/responses

​
Global selection

Set the default API surface for every request made through the provider: 
use AiSdk\Ollama;

Ollama::create(['api' => 'responses']);
There is no environment variable for this option — it must be set programmatically via Ollama::create().

​
Per-request override

Override the global default for a single request using providerOptions(). This does not affect other requests in the same process. 
use AiSdk\Generate;
use AiSdk\Ollama;

$result = Generate::text('Summarize this.')
    ->model(Ollama::model('llama3.2'))
    ->providerOptions('ollama', ['api' => 'responses'])
    ->run();
Per-request providerOptions take precedence over the global api option set in Ollama::create(). The request-level value is resolved by OllamaApi::resolve() each time, so the same validation rules apply.

​
OllamaApi enum

Internally, the provider maps string values to the OllamaApi enum via OllamaApi::resolve(). The two valid string values are:
  • chat_completions — maps to OllamaApi::ChatCompletions
  • responses — maps to OllamaApi::Responses
Resolution is case-insensitive and trims surrounding whitespace. Passing any other value — whether globally in Ollama::create() or as a per-request provider option — throws an InvalidArgumentException: 
Invalid Ollama API surface. Expected chat_completions or responses.

​
Restrictions on the Responses surface

The Responses API does not support reasoning controls. Passing .reasoning() on a request routed to responses throws:
InvalidArgumentException: Ollama Responses does not expose reasoning controls.
Use chat_completions or let the selected thinking model reason automatically.
Switch to chat_completions to use reasoning effort, or omit .reasoning() entirely and let the installed thinking model reason on its own.
The Responses API does not support structured output controls. Passing .output() on a request routed to responses throws:
InvalidArgumentException: Ollama Responses does not expose structured-output controls.
Use chat_completions for portable structured output.
Switch to chat_completions to use schema-constrained structured output.

Build docs developers (and LLMs) love

Get started for freeTalk to us