loading
loading
Automatically re-attempt a failed call, and if it keeps failing, switch to a backup path.
Retry re-runs a transient failure (timeout, rate limit) — ideally with backoff so you don't stampede — while fallback swaps to a different model, tool, or cached answer when the primary won't recover. Together they keep agents alive through the flakiness of real networks and APIs, but only safely on idempotent operations, or you'll double-charge a card. Example: a call to the big model 429s, retries twice with backoff, then falls back to a smaller model so the user still gets an answer.
Plainly
Think of Retry / Fallback as the checklist that keeps the app open for real people. Automatically re-attempt a failed call, and if it keeps failing, switch to a backup path.
In practice
Use it when a change has to survive deploys, users, incidents, analytics, or billing reality. In practice, define the owner, input, output, and failure mode before you rely on it.