Add a cache directive to the routing config to serve repeat requests without an upstream call. mode is required, simple (exact match) or semantic, with optional ttl_seconds, namespace, force_refresh, and max_response_bytes.
Cache-participating responses carry an x-routeplane-cache header so you always know what happened:
| x-routeplane-cache | Meaning |
|---|---|
hit | Served from cache, no upstream call. |
miss | Not cached; fetched upstream and stored. |
refreshed | Stale entry re-fetched and replaced. |
bypass | Caching skipped (e.g. force_refresh). |
Embeddings are never cached.