Struct OpenAiLanguageModel 

pub struct OpenAiLanguageModel { /* private fields */ }

OpenAI Chat Completions API language model.

Calls POST {base_url}/chat/completions. Works with OpenAI directly or any OpenAI-compatible endpoint via OpenAiConfig::base_url.

Compatible Providers

Provider	Base URL	Start command
OpenAI	https://api.openai.com/v1 (default)	—
MLX (text)	http://localhost:8080/v1	mlx_lm.server --model <model>
MLX (vision)	http://localhost:8080/v1	mlx_vlm.server --model <model>
vLLM	http://<host>/v1	vllm serve <model>

The Ollama daemon has its own provider impl (crate::OllamaLanguageModel) with reasoning-effort defaults that differ from OpenAI’s.

Implementations

impl OpenAiLanguageModel

pub fn new(deps: OpenAiDeps, config: OpenAiConfig) -> Self

Create a new OpenAI language model.

Trait Implementations

impl LanguageModelProvider for OpenAiLanguageModel

fn name(&self) -> &'static str

Stable provider key — used by the application router for model_id.split_once('/') routing. Examples: "anthropic", "openai", "ollama", "bedrock". Must be unique across the providers registered for one modality.

fn capabilities(&self, model: &ModelId) -> Option<ModelCapabilities>

Sync capability lookup against the provider’s static MODEL_CAPABILITIES table. Returns None when the model is not in the table (treat as conservative-deny for media content; pure text calls bypass this lookup).

fn list_models<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<Vec<ChatModelInfo>, LanguageModelError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Catalog of models this provider serves.

fn generate<'life0, 'life1, 'async_trait>( &'life0 self, request: GenerateRequest<'life1>, ) -> Pin<Box<dyn Future<Output = Result<LanguageModelResponse, LanguageModelError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Generate a complete response.

fn generate_stream<'life0, 'life1, 'async_trait>( &'life0 self, request: GenerateRequest<'life1>, ) -> Pin<Box<dyn Future<Output = Result<Pin<Box<dyn Stream<Item = Result<StreamDelta, LanguageModelError>> + Send>>, LanguageModelError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Generate a streamed response — token deltas as a Stream.

fn validate_request( &self, request: &GenerateRequest<'_>, ) -> Result<(), CapabilityError>

Validate that request only carries content parts the model actually accepts. Default impl walks request.messages, looks up the model in the provider’s capability table, and returns the first CapabilityError variant that applies — or Ok(()) for pure-text requests against text-only models.