Search by

symfony / ai-ollama-platform

fabpot

Ollama platform bridge for Symfony AI

Package info

github.com/symfony/ai-ollama-platform

Type:symfony-ai-platform

pkg:composer/symfony/ai-ollama-platform

Security

Advisories: 0

Aikido package health analysis

Transparency log

Fund package maintenance!

fabpot

nicolas-grekas

symfony.com/sponsor

Tidelift

Statistics

Installs: 312 569

Dependents: 13

Suggesters: 5

Stars: 9

v1.0.0-BETA1 2026-10-26 14:03 UTC

Requires

Requires (Dev)

Suggests

None

Provides

None

Conflicts

None

Replaces

None

MIT 19605f1b01b1a0549354b8ac4d1a86751bb1f101

  • Christopher Hertel <mail.woop@christopher-hertel.de>
  • Oskar Stark <oskarstark.woop@googlemail.com>
  • Symfony Community

Bridgeplatformlocalaiollama

Published on Packagist.org on 2026-10-26 14:08 UTC

This package is auto-updated.

Last update: 2026-11-03 20:45:42 UTC

README

Ollama platform bridge for Symfony AI.

Decision models

Models declaring the decision capability (Ollama >= 0.35.0) are served through POST /v1/systemone. This requires symfony/ai-decision-platform. They are queried with the Evaluation of the Decision bridge and answer with its Answers.

Clef and Clef Flash accept images on Ollama >= 0.35.1. Pass platform Image objects in the optional images argument of Evaluation. PNG, JPEG, and WebP images are sent as raw base64 data, in order, and shared by all questions. The state remains required. For Ollama cloud requests, the documented limits are at most 4 images, 4 MiB and 16 megapixels per image, and 8 MiB in total. Ollama's server enforces image formats and request limits; this bridge does not inspect image bytes or enforce those cloud limits locally.

The model catalog maps vision to Capability::INPUT_IMAGE for decision models. Images sent to a decision model without that capability are rejected by the bridge before the decision request is sent. Custom model catalogs must declare Capability::INPUT_IMAGE for image-capable decision models.

use Symfony\AI\Platform\Bridge\Decision\Evaluation;
use Symfony\AI\Platform\Bridge\Decision\Question\NoulQuestion;
use Symfony\AI\Platform\Message\Content\Image;

$result = $platform->invoke('clef-flash', new Evaluation(
    'A scanned yearbook photograph.',
    ['portrait' => new NoulQuestion('Is this a posed individual portrait?')],
    images: [Image::fromFile('/path/to/photo.jpg')],
));

Image URLs must be fetched by the application and converted to Image objects; the decision endpoint does not accept URLs or data URLs.

Text-to-Speech

Speech models such as phonon (Ollama >= 0.40.9) are served through the OpenAI-compatible POST /v1/audio/speech endpoint. Pass the text and, optionally, a voice, then write the resulting audio to a file:

use Symfony\AI\Platform\Bridge\Ollama\Factory;
use Symfony\AI\Platform\Bridge\Ollama\TextToSpeech\Format;

$platform = Factory::createPlatform('http://127.0.0.1:11434');

$result = $platform->invoke('phonon', 'Hello from Ollama.', [
    'voice' => 'Zoey',
    'response_format' => Format::WAV,
]);

$result->asFile('speech.wav');

Ollama answers with 16-bit mono audio at 24 kHz, in wav (default) or pcm format, for an input of up to 4,096 characters. Omitting voice uses the model's default voice. The API streams the audio while it is generated; the bridge buffers it into a binary result, so stream and stream_format are rejected locally. An unknown voice or another unsupported option is rejected by Ollama with a 400 error listing the available voices or supported values.

Ollama Documentation

Resources