Used by AIProviderBase to filter capability and route tasks. This provider does not advertise local/edge execution out of the box; if you need on-device models, use a provider that returns InferenceType.OnDevice, InferenceType.Cloud or InferenceType.LocalServer.

Signature

override InferenceType DefaultSupportedTypes

Properties

SupportsVision : bool

[Get]

Indicates whether this provider can handle vision inputs (images) alongside text during chat.

When true, ChatAsync will package ImageInput items using the OpenAI Responses format and, depending on settings, inline or resolve remote URLs. Toggle this for models like GPT-4o or any multimodal model that supports image understanding.

Controlled by the serialized supportsVision field and reported via IChatTask.SupportsVision. See also inlineRemoteImages and resolveRemoteRedirects to influence how remote images are prepared.

Signature

bool SupportsVision

Methods

ChatAsync ( req , stream , ct )	Sends a chat turn to OpenAI's Responses API and returns the assistant's text reply. Validates `apiKey` and `model`, builds a single `input` message per the Responses schema, and POSTs to `{apiRoot}/v1/responses`. The method extracts `output_text` if present, or falls back to the first text item in the first output message. OpenAI Responses docs: https://platform.openai.com/docs/guides/responses/⁠ See alsoAIProviderBaseIChatTask. Signature async Task< ChatResponse > ChatAsync(ChatRequest req, IProgress< ChatDelta > stream=null, CancellationToken ct=default) Parameters `req: ChatRequest` The user message and optional ImageInput list. If SupportsVision is enabled, images are serialized as `input_image` items; remote images can be inlined or have redirects resolved depending on inlineRemoteImages and resolveRemoteRedirects. `stream: IProgress< ChatDelta >` Optional incremental callback for partial text via ChatDelta. This implementation reports the final text once per call (non-streaming HTTP). Use to update UI progressively. `ct: CancellationToken` Cancellation token for image preparation and HTTP. Cancels the request if the operation is aborted. Returns `async Task< ChatResponse >` A ChatResponse containing the assistant text and the raw JSON payload for debugging or downstream parsing.
SynthesizeStreamCoroutine ( text , voice , onReady )	Coroutine that synthesizes speech with OpenAI audio/speech and yields a Unity AudioClip. Selects AudioType based on ttsOutputFormat (e.g., WAV/MP3). Requires valid `apiKey` and `model`. POSTs to `{apiRoot}/v1/audio/speech`, then streams the response into a AudioClip via the internal HTTP helper. OpenAI TTS: https://platform.openai.com/docs/guides/text-to-speech⁠ See also ITextToSpeechTask. Signature IEnumerator SynthesizeStreamCoroutine(string text, string voice=null, Action< AudioClip > onReady=null) Parameters `text: string` Input text to speak. Logs and exits if empty. Combined with ttsVoice and optional ttsInstructions to control style, plus ttsSpeed for playback rate. `voice: string` Optional voice name override. If null/empty, uses ttsVoice. `onReady: Action< AudioClip >` Callback invoked with the created AudioClip once download/decoding completes. Returns `IEnumerator`
TranscribeAsync ( audioBytes , language , ct )	Transcribes an audio clip using OpenAI audio/transcriptions and returns plain text. Honors sttResponseFormat (e.g., `json` or `text`) and optional sttTemperature. Requires valid `apiKey` and `model`. POSTs to `{apiRoot}/v1/audio/transcriptions`. OpenAI Transcriptions: https://platform.openai.com/docs/guides/speech-to-text⁠ See also ISpeechToTextTask. Signature async Task< string > TranscribeAsync(byte[] audioBytes, string language=null, CancellationToken ct=default) Parameters `audioBytes: byte[]` Raw audio data (e.g., WAV). Throws if null or empty. The content is sent as multipart/form-data. `language: string` Optional ISO language override (for example, `"en"`, `"de"`). If null/empty, falls back to sttLanguage or lets OpenAI auto-detect. `ct: CancellationToken` Cancellation token for the HTTP request. Returns `async Task< string >` Transcript text. If sttResponseFormat is `"text"`, the raw body is returned. Throws `ArgumentException` Thrown when audioBytes is null or empty. `InvalidOperationException` Thrown if `apiKey` or `model` is missing.