Expand description
Transient-error retry wrapper around a CompletionModel.
RetryingModel wraps the OpenAi-backed completion model so that each
individual model call retries on transient HTTP failures (request timeouts,
connection errors, 408/425/429/5xx) with bounded exponential backoff.
Retrying at the model-call layer – rather than around agent.prompt(...)
– is deliberate: a single user turn drives a model -> tool -> model loop,
and wrapping the whole prompt would re-execute already-run container tool
calls when a later model call fails. A failed model call has produced no
tool calls yet, so retrying just that call replays nothing observable.
Structs§
- Retrying
Model - Wraps a
CompletionModel, retrying transient failures on every call.