Struct AiClientBuilder 

pub struct AiClientBuilder { /* private fields */ }

Builder for creating clients with custom configuration.

Keep this surface area small and predictable (developer-friendly).

Sharing across tasks

AiClient does not implement Clone (by design, for API key and ToS compliance). To share a client across multiple async tasks, wrap it in Arc:

let client = Arc::new(
    AiClientBuilder::new()
        .build("openai/gpt-4o")
        .await?
);
// Use Arc::clone(&client) to pass to tasks
tokio::spawn(use_client(Arc::clone(&client)));

Implementations

impl AiClientBuilder

pub fn new() -> AiClientBuilder

pub fn protocol_path(self, path: String) -> AiClientBuilder

Set custom protocol directory path.

pub fn hot_reload(self, enable: bool) -> AiClientBuilder

Enable hot reload of protocol files.

pub fn with_fallbacks(self, fallbacks: Vec<String>) -> AiClientBuilder

Set fallback models.

pub fn strict_streaming(self, enable: bool) -> AiClientBuilder

Enable strict streaming validation (fail fast when streaming config is incomplete).

This is intentionally opt-in to preserve compatibility with partial manifests.

pub fn feedback_sink(self, sink: Arc<dyn FeedbackSink>) -> AiClientBuilder

Inject a feedback sink. Default is a no-op sink.

pub fn max_inflight(self, n: usize) -> AiClientBuilder

Limit maximum number of in-flight requests/streams. This is a simple backpressure mechanism for production safety.

pub fn base_url_override(self, base_url: impl Into<String>) -> AiClientBuilder

Override the base URL from the protocol manifest.

This is primarily for testing with mock servers. In production, use the base_url defined in the protocol manifest.

pub async fn build(self, model: &str) -> Result<AiClient, Error>

Build the client.

Trait Implementations

impl Default for AiClientBuilder

fn default() -> AiClientBuilder

Returns the “default value” for a type.