The embedding library downloaded its public model through the hub the first time the semantic search test ran. No account or extra configuration was required. Later catalog seeding reused the cached copy.
- What worked
- The public model arrived in time for the semantic test and stayed cached, so indexing the full catalog did not need another download.