Struct AiClient 

pub struct AiClient {
    pub manifest: ProtocolManifest,
    pub transport: Arc<HttpTransport>,
    pub pipeline: Arc<Pipeline>,
    pub loader: Arc<ProtocolLoader>,
    /* private fields */
}

Unified AI client that works with any provider through protocol configuration.

Fields

manifest: ProtocolManifest

transport: Arc<HttpTransport>

pipeline: Arc<Pipeline>

loader: Arc<ProtocolLoader>

Implementations

impl AiClient

pub async fn signals(&self) -> SignalsSnapshot

Snapshot current runtime signals (facts only) for application-layer orchestration.

pub async fn new(model: &str) -> Result<Self>

Create a new client for a specific model.

The model identifier should be in the format “provider/model-name” (e.g., “anthropic/claude-3-5-sonnet”)

pub fn chat(&self) -> ChatRequestBuilder<'_>

Create a chat request builder.

pub async fn chat_batch( &self, requests: Vec<ChatBatchRequest>, concurrency_limit: Option<usize>, ) -> Vec<Result<UnifiedResponse>>

Execute multiple chat requests concurrently with an optional concurrency limit.

Notes:

• Results preserve input order.
• Internally uses the same “streaming → UnifiedResponse” path for consistency.

pub async fn chat_batch_smart( &self, requests: Vec<ChatBatchRequest>, ) -> Vec<Result<UnifiedResponse>>

Smart batch execution with a conservative, developer-friendly default heuristic.

• For very small batches, run sequentially to reduce overhead.
• For larger batches, run with a bounded concurrency.

You can override the chosen concurrency via env:

• AI_LIB_BATCH_CONCURRENCY

pub async fn report_feedback(&self, event: FeedbackEvent) -> Result<()>

Report user feedback (optional). This delegates to the injected FeedbackSink.

pub async fn update_rate_limits(&self, headers: &HeaderMap)

Update rate limiter state from response headers using protocol-mapped names.

This method is public for testing purposes.

pub async fn call_model( &self, request: UnifiedRequest, ) -> Result<UnifiedResponse>

Unified entry point for calling a model. Handles text, streaming, and error fallback automatically.

pub async fn call_model_with_stats( &self, request: UnifiedRequest, ) -> Result<(UnifiedResponse, CallStats)>

Call a model and also return per-call stats (latency, retries, request ids, endpoint, usage, etc.).

This is intended for higher-level model selection and observability. Call a model and also return per-call stats (latency, retries, request ids, endpoint, usage, etc.).

This is intended for higher-level model selection and observability.

pub fn validate_request(&self, request: &ChatRequestBuilder<'_>) -> Result<()>

Validate request capabilities.

Trait Implementations

impl EndpointExt for AiClient

async fn call_service(&self, service_name: &str) -> Result<Value>

Call a generic service by name.

async fn list_remote_models(&self) -> Result<Vec<String>>

List models available from the provider.

fn resolve_endpoint(&self, name: &str) -> Result<&EndpointConfig>