public/api/types/options library

Classes

AudioFormat
Container format of an audio payload.
BackendPreference
One ranked backend choice for LoadOptions.backendPreferences.
DiarizationOptions
Clustering controls for diarization.
EmbeddingsPoolingStrategy
The required public spelling in every SDK is exactly "mean" / "cls" / "last". LAST is the final token's hidden state (llama.cpp --pooling last), which decoder-style embedders require. It is NOT max-pooling; no SDK may expose it as "max".
EmbedOptions
Normalization and pooling controls for embeddings.
Endpointing
Silence window that ends a user turn in a voice session.
ImageMode
Whether images.generate synthesizes from scratch or repaints a region.
ImageOptions
Size, sampling, and progress controls for images.
InferenceFramework
Engine that executes a model. Reached from ModelInfo.framework, so the reserved values below are manifest-critical.
Interruption
Barge-in policy for a voice session.
LlmOptions
Sampling, prompting, reasoning, and tool controls for llm and vlm.
LoadOptions
Placement knobs applied when a model is loaded.
ModelCategory
What a model does.
RagConfig
Chunking, retrieval, and persistence controls for a RAG session.
RagQueryOptions
Per-query knobs for RagSession.query/queryStream.
RagRetrievalOptions
Per-query retrieval overrides for RagSession.query/queryStream.
ReasoningMode
The single home for reasoning control. Composed surfaces (tool calling, RAG, voice agent) inherit it through the embedded LLMGenerationOptions.
ReasoningOptions
Thinking-phase control for a generation request.
SegmentationOptions
Diagnostic controls for segmentation.
StructuredOutput
Schema constraint applied to a generation's output.
SttOptions
Language, formatting, and speaker controls for stt.
ToolChoice
How the model may use registered tools during a generation.
TtsOptions
Voice, prosody, and output-format controls for tts.
TurnHandlingOptions
Turn-taking behaviour of a voice session.
VadOptions
Sensitivity and timing controls for vad.

Enums

AcceleratorPolicy
Hardware class LoadOptions.accelerator requests.
StructuredOutputMode
Enforcement level llm.generateStructured applies to its schema.