Struct OpenAiResponsesAdapter 

pub struct OpenAiResponsesAdapter { /* private fields */ }

Live OpenAI Responses API adapter.

It implements ProviderAdapter and delegates all runtime policy, journaling, event publication, approval, and tool execution back to agent-sdk-core.

Implementations

impl OpenAiResponsesAdapter

pub fn from_env(model: impl Into<String>) -> Result<Self, AgentError>

Creates a live adapter using OPENAI_API_KEY.

pub fn new( config: OpenAiLiveResponsesConfig, api_key: ProviderApiKey, ) -> Result<Self, AgentError>

Creates a live adapter with a host-resolved API key.

pub fn with_transport( config: OpenAiLiveResponsesConfig, api_key: ProviderApiKey, http: Arc<dyn JsonHttpTransport>, ) -> Result<Self, AgentError>

Creates an adapter with an injected JSON transport for deterministic tests or host-managed HTTP stacks.

pub fn with_argument_sink(self, sink: Arc<dyn OpenAiToolArgumentSink>) -> Self

Adds an optional host-owned sink for raw tool-call arguments.

Trait Implementations

impl Clone for OpenAiResponsesAdapter

fn clone(&self) -> OpenAiResponsesAdapter

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl ProviderAdapter for OpenAiResponsesAdapter

fn capabilities(&self) -> ProviderCapabilities

Returns adapter capability metadata for policy and package resolution. This is data-only and does not perform I/O, call host ports, append journals, publish events, or start processes.

fn project_request( &self, projection: &ContextProjection, policy: &ProviderProjectionPolicy, ) -> Result<ProviderRequest, AgentError>

Projects admitted context into the provider’s request shape. This projects admitted context into a provider request and must not fetch hidden raw content.

fn complete( &self, request: &ProviderRequest, ) -> Result<ProviderResponse, AgentError>

Calls the provider for one non-streaming completion request. Implementations may call the model provider; caller-owned runtime code must handle policy, journaling, and event publication around it.

fn stream( &self, request: &ProviderRequest, ) -> Result<Vec<ProviderStreamChunk>, AgentError>

Calls the provider for a streaming response. Implementations may call the model provider; caller-owned runtime code must handle policy, journaling, and event publication around it.

fn extract_usage(&self, response: &ProviderResponse) -> ProviderUsage

Extracts provider usage accounting from a response. This derives usage accounting from a provider response and performs no provider call.