Struct OpenAIResponsesClient 

pub struct OpenAIResponsesClient { /* private fields */ }

Available on crate features models and openai only.

Client for the OpenAI Responses API (/responses endpoint).

Wraps async-openai’s typed Responses client and implements adk_core::Llm. Supports reasoning summaries, conversation state via previous_response_id, and built-in tools (web search, file search, code interpreter).

Example

use adk_model::openai::{OpenAIResponsesClient, OpenAIResponsesConfig};

let config = OpenAIResponsesConfig::new("sk-...", "o3");
let client = OpenAIResponsesClient::new(config)?;

Implementations

impl OpenAIResponsesClient

pub async fn poll_response( &self, response_id: &str, ) -> Result<LlmResponse, AdkError>

Poll a background response by ID.

Calls GET /v1/responses/{response_id} and returns the current status or completed result as an LlmResponse.

Status Handling

• Completed: Returns a full LlmResponse with content, usage, and metadata.
• In Progress / Queued: Returns an LlmResponse with provider_metadata containing response_id and status, but no content.
• Failed: Returns an LlmResponse with error_code and error_message populated from the response error object.

Errors

Returns AdkError if the HTTP request fails (network error, auth error, etc.).

Example

let response = client.poll_response("resp_abc123").await?;
if let Some(metadata) = &response.provider_metadata {
    let status = metadata["openai"]["status"].as_str();
    println!("Response status: {:?}", status);
}

pub async fn cancel_response( &self, response_id: &str, ) -> Result<LlmResponse, AdkError>

Cancel a background response by ID.

Calls POST /v1/responses/{response_id}/cancel to cancel a response that was created with background: true. Only background responses can be cancelled.

Returns

Returns an LlmResponse with status: "cancelled" in provider_metadata on success.

Errors

Returns AdkError if:

• The HTTP request fails (network error, auth error)
• The response is not a background response
• The response has already completed

Example

let response = client.cancel_response("resp_abc123").await?;
println!("Response cancelled: {:?}", response.provider_metadata);

impl OpenAIResponsesClient

pub async fn compact_response( &self, response_id: &str, ) -> Result<LlmResponse, AdkError>

Explicitly compact a conversation by response ID.

Calls POST /v1/responses/{id}/compact and returns the compacted response. The compacted response ID is preserved in provider_metadata["openai"]["response_id"] for subsequent previous_response_id chaining.

Errors

Returns AdkError with category NotFound and code model.openai_responses.compaction_not_found if the response ID does not exist (404).

Example

let compacted = client.compact_response("resp_abc123").await?;
let new_id = compacted.provider_metadata
    .as_ref()
    .and_then(|m| m.get("openai"))
    .and_then(|o| o.get("response_id"))
    .and_then(|v| v.as_str());

impl OpenAIResponsesClient

pub fn new( config: OpenAIResponsesConfig, ) -> Result<OpenAIResponsesClient, AdkError>

Create a new Responses API client from the given config.

Errors

Returns AdkError with InvalidInput if api_key is empty and Open Responses mode is not enabled with a custom base URL.

pub fn new_with_reasoning_effort( config: OpenAIResponsesConfig, reasoning_effort: OpenAIReasoningEffort, ) -> Result<OpenAIResponsesClient, AdkError>

Create a Responses API client with the complete reasoning-effort vocabulary.

This constructor supports the none, minimal, xhigh, and max values used across current OpenAI model generations while retaining the original configuration field for backward compatibility.

pub fn with_retry_config( self, retry_config: RetryConfig, ) -> OpenAIResponsesClient

Set the retry configuration, consuming self.

pub fn set_retry_config(&mut self, retry_config: RetryConfig)

Set the retry configuration by mutable reference.

pub fn retry_config(&self) -> &RetryConfig

Get a reference to the current retry configuration.

pub fn is_open_responses_mode(&self) -> bool

Whether Open Responses mode is enabled.

When true, the client relaxes strict OpenAI field validation for compatibility with third-party Open Responses-compatible endpoints.

Trait Implementations

impl Llm for OpenAIResponsesClient

fn name(&self) -> &str

Returns the model identifier (e.g., “gemini-2.5-flash”).

fn generate_content<'life0, 'async_trait>( &'life0 self, request: LlmRequest, stream: bool, ) -> Pin<Box<dyn Future<Output = Result<Pin<Box<dyn Stream<Item = Result<LlmResponse, AdkError>> + Send>>, AdkError>> + Send + 'async_trait>>

where 'life0: 'async_trait, OpenAIResponsesClient: 'async_trait,

Generates content from the given request, optionally streaming.

fn schema_adapter(&self) -> &dyn SchemaAdapter

Returns the schema adapter for this provider.

fn uses_interactions_api(&self) -> bool

Returns true if this model is configured to use a server-managed environment (e.g., Gemini Interactions API) where the provider owns the tool-calling loop and filesystem.