Optionalconfig: AIFactoryConfigProtecteddefaultsOptionaloptions: Partial<AIRequest>All model IDs supported by any registered provider.
First provider that lists the given model, or null.
One answer, with retries and fallback.
A retryable failure (rate limit, overload, network, timeout) is retried
with backoff, honouring Retry-After; anything else moves straight to the
next fallback provider. Never throws for a provider failure: the result
carries success: false and a classified errorInfo.
The same call as [process], streamed.
Retry and fallback apply only until the first chunk arrives: by then the
caller has usually shown part of an answer, and restarting on another
provider would splice two different answers together. A failure after
that is thrown from the iterator as an AIError.
Runs provider discovery if it has not run yet.
Test each registered provider; returns map of providerId -> ok.
Providers used when the config names none. Empty here; the
.entry's subclass supplies the built-ins.