Module retry 

Exponential backoff retry wrapper for LLMClient. Exponential backoff retry wrapper for LLMClient.

Wraps any LLMClient implementation with configurable retry logic. Retries on rate-limit (429) and server errors (5xx) with exponential backoff and jitter to avoid thundering-herd behavior.

Streaming is not retried — the caller receives the raw stream from the inner client. Streaming retry is complex and left for a future version.

Example

use llm_kernel::llm::{LLMClient, OpenAIClient, RetryClient, RetryConfig};
use std::time::Duration;

let client = OpenAIClient::from_key("gpt-4o", "sk-...")?;
let retry = RetryClient::new(client, RetryConfig {
    max_retries: 3,
    base_delay: Duration::from_secs(1),
});
let response = retry.complete(request).await?;

Structs

RetryClient

An LLMClient wrapper that retries on transient errors.

RetryConfig

Configuration for retry behavior.