Every query template names the repository it runs against, as the object of <https://w3id.org/kpxl/grlc/endpoint> in its assertion. The usual choice is the full repository, which holds every loaded nanopublication:

<https://w3id.org/np/l/nanopub-query-1.1/repo/full>

Other repositories are narrower and therefore faster: meta for nanopublication metadata, text for full-text search, last30d for recent nanopublications, and a per-type repository type/<hash> or per-key repository pubkey/<hash>. Two further repositories, spaces and trust, hold the server-side computed state of space membership and of the agent-to-key trust network; a query can reach them from another repository with a federated service clause.

Always write the endpoint with the generic w3id.org/np/l/nanopub-query-1.1/ prefix rather than the host name of one instance. The prefix is rewritten to whichever instance answers the query, so a template naming a single host stops working as soon as that host does.